diff --git a/docs/evidence.md b/docs/evidence.md index 40556bc8b..dd67e7105 100644 --- a/docs/evidence.md +++ b/docs/evidence.md @@ -21,7 +21,7 @@ Four rules for reading the table: 4. The 12-node cloud network of 4 October 2026 (`infra/cloud-devnet`, cloud VMs in five locations) is the project's own. Rows that cite it are tested by the team, not reproduced externally. 5. The labels stay distinct and a claim never moves up a label without the artefact that label names (the table "What would move a row"). The public reference repository exists now (the specifications, the `igneum-pow` crate, the simulators and the test material, at git.igneum.network); the full node, the miner and the proving code open later, so a row that cites them is tested by the team until they do. -Versions in the table: `igneum-pow` is the Rust crate at `igneum-pow/Cargo.toml`, version 0.2.0 since 4 October 2026 (generator version 2; 0.1.0 rows are marked). "Repo" commits are this repository's. "Fork" commits are `vendor/igneum-node` and its worktrees (`-v4`, `-diff`, `-exec`, `-harness`, `-fin-fixes`), which are not in this repository's history; the row names the fork commit or branch as the bench log does. The first devnet is devnet v4 (genesis 10:05 BST, 4 October 2026, branch `devnet-v4`). Devnet 3 (`igneum-devnet-3`, chain id 4463 from genesis and {{rm:network.chain_id}} since its class v5 floor at DAA 68,400; node {{rm:source.node.commit}} on {{rm:source.node.branch}}, igneum-pow {{rm:source.pow.commit}}, {{rm:read_at_short}}; the current state is the [release manifest](/release.json), filled at build time) made its first block at 17:06 UTC on 7 October 2026 with every upgrade on from block zero (class v4 sub-version 3, the era VDF, difficulty v2, finality v3, proving v0 and v1, the ladder at rung 0, calibrated fees) and locked its first checkpoint at 19:02 UTC; rows that name the live devnet without a chain are the first devnet's, and Devnet 3 readings are marked. The spec is `docs/spec/` version 0.1. +Versions in the table: `igneum-pow` is the Rust crate at `igneum-pow/Cargo.toml`, version 0.2.0 since 4 October 2026 (generator version 2; 0.1.0 rows are marked). "Repo" commits are this repository's. "Fork" commits are `vendor/igneum-node` and its worktrees (`-v4`, `-diff`, `-exec`, `-harness`, `-fin-fixes`), which are not in this repository's history; the row names the fork commit or branch as the bench log does. The first devnet is devnet v4 (genesis 10:05 BST, 4 October 2026, branch `devnet-v4`). The devnet ( chain id 4463 from genesis and {{rm:network.chain_id}} since its class v5 floor at DAA 68,400; node {{rm:source.node.commit}} on {{rm:source.node.branch}}, igneum-pow {{rm:source.pow.commit}}, {{rm:read_at_short}}; the current state is the [release manifest](/release.json), filled at build time) made its first block at 17:06 UTC on 7 October 2026 with every upgrade on from block zero (class v4 sub-version 3, the era VDF, difficulty v2, finality v3, proving v0 and v1, the ladder at rung 0, calibrated fees) and locked its first checkpoint at 19:02 UTC; rows that name the live devnet without a chain are the first devnet's, and the devnet readings are marked. The spec is `docs/spec/` version 0.1. ## The table @@ -33,17 +33,17 @@ Versions in the table: `igneum-pow` is the Rust crate at `igneum-pow/Cargo.toml` | 4 | The same program produces identical hashes on three GPU vendors, cache and dataset included | Litepaper vs RandomX ("Bit-exact on Apple, NVIDIA and AMD, measured"), For miners; homepage | tested by the team | repo `f2e903e`, `0f1fdaf` (version 1 packs), `b27da39` (version 2 packs `igneum-genesis-mh`, `igneum-devnet-v4-epoch0`); igneum-pow 0.2.0 | The 96 test vectors of a pack through `proto-metal/igneum-bench`, `proto-cuda/host.cu`, `proto-opencl/host.c`; batch fingerprint at `--batch-log2 24`; the miner's CPU re-check of every share a GPU worker finds on the devnet; bench-log entries "RTX 5090, memory-hard dataset", "AMD gfx1036", "RTX 5090 through NVIDIA OpenCL", "generator version 2 adopted", "the gfx1036 worker fault" | Version 1: 96/96 on Apple Metal (M5 Max), NVIDIA CUDA and NVIDIA OpenCL (RTX 5090, Windows), AMD OpenCL (Ryzen 7 9800X3D integrated gfx1036, 1 compute unit), Apple OpenCL, pocl and two CPU references; batch fingerprint `98af644e993239e2` over 16.7 million nonces identical on the AMD chip and the 5090, 3 October 2026. Version 2: 96/96 on Apple Metal, Apple OpenCL and the CUDA and OpenCL emulators with identical fingerprints; on real NVIDIA and AMD silicon the version 2 vectors have not run as a pack, but both mined accepted blocks on the live devnet with the CPU re-check clean on every share (RTX 5090 at 124.2 MH/s, gfx1036 at 3.3 MH/s), 4 October 2026. The AMD device is an integrated chip; no discrete AMD card and no Intel card has run anything (O-1.15) | none yet | | 5 | A CPU verifies one hash in under 10 ms by simulating one warp | Litepaper Mining ("about ten milliseconds"), vs RandomX; roadmap gate 2 | tested by the team | repo `75cac18`, `b27da39`; igneum-pow 0.2.0 (`verify.rs`) | `cargo test` and the crate bench in `igneum-pow/`; bench-log "igneum-pow: Rust crate bit-exact with proto-metal" and "generator version 2 adopted" | 0.411 to 0.579 ms per 32-lane warp steady, 0.41 to 0.87 ms cold, average of 20, 1 GiB dataset, cache held, one M5 Max performance core, 3 October 2026; version 2 units 0.631 ms (average of 20), cold 0.67 to 0.81 ms, 4 October 2026. Gate margin about 16x on this core. Not measured on a 2019-class laptop core (O-1.14) | none yet | | 6 | The hash is bound to the header: one nonce serves one header, and a wrong nonce is rejected | Spec 1.6; litepaper Mining (implied by "checks a hash") | tested by the team | repo `33f7b33`, `9812466`, `b27da39`; igneum-pow 0.2.0 (`bind.rs`, bound vectors re-cut for version 2, 39 crate tests) | `igneum-miner bad-nonce` against a devnet node; `igneum-pow hash-bound` for the 96-nonce job across the 2^32 lane boundary; bench-log "first devnet blocks on the real lottery hash" and "generator version 2 adopted" | 833 blocks accepted by `igneum-lottery-v1-bound` on 3 nodes, 0 rejections; `bad-nonce` gave Reject(BlockInvalid); Metal, OpenCL and CUDA (emulated) workers bit-exact with the crate on the lane-boundary job, 3 October 2026, Apple M5 Max. Version 2: the node's engine reports `igneum-lottery-v2-bound`, 39 of 39 crate tests, and the live devnet v4 accepts its blocks under it, 4 October 2026 | none yet | -| 7 | The devnet runs at one block a second | Homepage stats ("1 / s"); litepaper Speed; roadmap phase 3 | tested by the team | repo `9812466`, `e9328c6`, `8dae48b`; fork `devnet-v4` `dc749905` | The merged node's 3-node test network (`igneum-devnet-880`, 960 s); the live devnet v4 record `sim/difficulty/records/live-2026-10-04.csv`; the 12-node cloud network's arrival logs; bench-log "devnet-v4 integration", "difficulty rule v2", "first devnet blocks" | Devnet 3, 7 October 2026: first block accepted at 17:06 UTC, 85 of 85 GPU blocks by 17:10 UTC and 287 by 17:14 UTC, 0 rejected (`docs/plans/release-0.3.22.md` section 5). Merged node, 4 October 2026, Apple M5 Max: 1,055 blocks in 960 s, 1.03 blocks/s, sink identical on 3 nodes at 31 of 31 samples, 0 rejected. Live devnet v4 the same day: 49 to 81 blocks a minute while two RTX 5090s joined and left (row 12), 1.1 to 1.2 blocks/s in the oscillating window, then within 1.3% per minute with one PC and the Mac. The 12-node cloud network at one block a second: 644 blocks in a 10-minute window. The 3 October CPU devnet: 1.29 blocks/s over 641 s, 1.03 after the first retarget. The phase 3 gate also asks for proofs under 60 s behind the tip; no proof is on the chain (row 15) | none yet | +| 7 | The devnet runs at one block a second | Homepage stats ("1 / s"); litepaper Speed; roadmap phase 3 | tested by the team | repo `9812466`, `e9328c6`, `8dae48b`; fork `devnet-v4` `dc749905` | The merged node's 3-node test network (`igneum-devnet-880`, 960 s); the live devnet v4 record `sim/difficulty/records/live-2026-10-04.csv`; the 12-node cloud network's arrival logs; bench-log "devnet-v4 integration", "difficulty rule v2", "first devnet blocks" | the devnet, 7 October 2026: first block accepted at 17:06 UTC, 85 of 85 GPU blocks by 17:10 UTC and 287 by 17:14 UTC, 0 rejected (`docs/plans/release-0.3.22.md` section 5). Merged node, 4 October 2026, Apple M5 Max: 1,055 blocks in 960 s, 1.03 blocks/s, sink identical on 3 nodes at 31 of 31 samples, 0 rejected. Live devnet v4 the same day: 49 to 81 blocks a minute while two RTX 5090s joined and left (row 12), 1.1 to 1.2 blocks/s in the oscillating window, then within 1.3% per minute with one PC and the Mac. The 12-node cloud network at one block a second: 644 blocks in a 10-minute window. The 3 October CPU devnet: 1.29 blocks/s over 641 s, 1.03 after the first retarget. The phase 3 gate also asks for proofs under 60 s behind the tip; no proof is on the chain (row 15) | none yet | | 8 | Blocks are mined by GPUs on Apple and NVIDIA | Homepage live strip; journey phase 3 ("GPU miners on three vendors") | tested by the team | repo `9812466`, `e9328c6`, `d7e1f89`, `2309c8d`; fork `devnet-v4` | Metal worker `proto-metal/igneum-bench --serve` driven by `igneum-miner --worker`; the live devnet v4 hash-rate record `sim/difficulty/records/live-2026-10-04-hashrate.csv` (587 worker STATUS lines by run id); bench-log "first devnet blocks", "devnet v4 cut-over", "difficulty rule v2", "first machine on the Igneum Miner app" | Metal: 506 jobs, 5,636 blocks found and accepted, 0 rejected, 0 CPU/GPU mismatches, 28.2 MH/s wall, 3 October 2026. Live devnet v4, 4 October 2026: PC 1's RTX 5090 at 122 MH/s with 8 identities, PC 2's at 124 MH/s with 8 identities (117 to 119 MH/s inside the one-click app, 34 accepted blocks in its first minute, CPU re-check OK on every share), the Mac's Metal worker at 26.7 MH/s; 17 vote keys signed the first finality lock (row 10); from the afternoon an Apple silicon laptop outside the project at 21.0 MH/s through the app (row 30). Two RTX 5090s and two Apple chips; no other NVIDIA model has mined | none yet | | 9 | Blocks are mined by a GPU on AMD | Journey phase 3 ("three vendors") | tested by the team | repo `2c4b30f` (generic OpenCL worker, `--pack`), `112acf6` (fault guards); bound kernel `kernel_bound.cl` in the pack | `igneum-worker-opencl.exe --pack` on PC 2's integrated Radeon against the live devnet v4 through the Windows package; bench-log "the gfx1036 worker fault", "first hourly program swap", "first machine on the Igneum Miner app" | PC 2's integrated gfx1036 (1 compute unit) mined on the live devnet on 4 October 2026: 8 accepted blocks at 3.3 MH/s over 577 s with the CPU re-check clean, and 2.74 MH/s through the hourly program swap with 0 rejected. At about 600 s the AMD runtime began answering every call with success while running nothing (906 jobs became 56,384 in 30 s, 4.3 GH/s of phantom work); not reproduced on Apple OpenCL in 4,565 jobs with 0 leaked objects; the worker and miner now refuse a job 20x faster than the mean or an unchanged output buffer and restart (`112acf6`), and the next gfx1036 run names the guard that fires. One integrated chip; no discrete AMD card has run anything | none yet | -| 10 | Checkpoints lock every 30 s of chain at two thirds of all 30-day weight, and the floor stops conflicting locks in partitions and eclipses for as long as neither side's own new blocks carry it past two thirds of its window (about 10 days of a 30-day window at a 50/50 split) | Litepaper Finality, "What Igneum does not claim"; homepage "locked every 30 seconds" | tested by the team | repo `a3a9833` (2/3 floor, O-3.15), `bbb264a` (simulation), `c16ccf1`; fork `devnet-v4` `6457ca95` (`FLOOR_NUM / FLOOR_DEN` 2/3), `da1eb889` (F17 by-weight sortition, F1 first-month gate `min_daa = window`); spec 3.3, 3.3.1, 3.7, 3.9 | The live devnet v4 (`getFinalityCheckpoints`, `tools/observer/observer.mjs`, `/api/checkpoint`); `sim/finality_v2.py --floor 1.0`, scenarios A to L; the three-node, six-voter partition runs `igneum-devnet-921` to `-923`; `tools/finality-attacks` scenarios 1 to 6 and 8; bench-log "first finality lock on the live devnet", "finality floor 2/3", "finality v2 attack harness", "finality fixes F17 and F1" | Devnet 3, 7 October 2026: finality rule v3 from block zero, first lock at 19:02 UTC, under two hours after the first block (`docs/plans/release-0.3.22.md`). The first devnet: the first lock on the live devnet was checkpoint 242 at 12:03:44 BST on 4 October 2026, two hours after genesis (the window and `min_daa` are 7,200 DAA), with 77.4% of all weight and of active weight signed by 12 aggregated votes from 17 vote keys; `observer.mjs` saw it 0.7 s after the miner's own lock line. By 13:21 UTC the observer held 280 certificates, indices 241 to 522 (DAA 7,229 to 17,982), 17 to 27 voters, no index with two hashes. Test networks, 4 October 2026, Apple M5 Max: a 4/2 split locked on the 4 side (67.9%) 2 to 8 s after the cut and never on the 2 side, 0 conflicts; a 3/3 split locked on neither side for 150 s with 0 conflicts, where the 3 October floor (56.7%) would have locked both sides at 76 and 106 s; the rule guarantees one lock history for partitions shorter than the window bound W / (3R) (200 s on that test network's 1,800-DAA window, about 40 minutes on the devnet, about 10 days at the 30-day mainnet window); beyond that bound each side can reach two thirds of its own window, so the next finality rule freezes the weight table at the last certified checkpoint and pauses instead. Simulator with the 2/3 floor: 0 conflicts up to a 33% equivocator (34% splits a 50/50 partition), silent weight pauses locks from 34%, a 50/50 partition locks alone from day 10.1. Harness: equivocating keys stripped on every node, Sybil dust at zero weight, a pulsed miner's weight equal to its block share (ratio 0.96 to 1.0), the first-month gate stops a young window locking under one key. Not demonstrated: certificate injection on the wire, an eclipse with a private fork, the 2-hour presence window at mainnet length | none yet | +| 10 | Checkpoints lock every 30 s of chain at two thirds of all 30-day weight, and the floor stops conflicting locks in partitions and eclipses for as long as neither side's own new blocks carry it past two thirds of its window (about 10 days of a 30-day window at a 50/50 split) | Litepaper Finality, "What Igneum does not claim"; homepage "locked every 30 seconds" | tested by the team | repo `a3a9833` (2/3 floor, O-3.15), `bbb264a` (simulation), `c16ccf1`; fork `devnet-v4` `6457ca95` (`FLOOR_NUM / FLOOR_DEN` 2/3), `da1eb889` (F17 by-weight sortition, F1 first-month gate `min_daa = window`); spec 3.3, 3.3.1, 3.7, 3.9 | The live devnet v4 (`getFinalityCheckpoints`, `tools/observer/observer.mjs`, `/api/checkpoint`); `sim/finality_v2.py --floor 1.0`, scenarios A to L; the three-node, six-voter partition runs `igneum-devnet-921` to `-923`; `tools/finality-attacks` scenarios 1 to 6 and 8; bench-log "first finality lock on the live devnet", "finality floor 2/3", "finality v2 attack harness", "finality fixes F17 and F1" | the devnet, 7 October 2026: finality rule v3 from block zero, first lock at 19:02 UTC, under two hours after the first block (`docs/plans/release-0.3.22.md`). The first devnet: the first lock on the live devnet was checkpoint 242 at 12:03:44 BST on 4 October 2026, two hours after genesis (the window and `min_daa` are 7,200 DAA), with 77.4% of all weight and of active weight signed by 12 aggregated votes from 17 vote keys; `observer.mjs` saw it 0.7 s after the miner's own lock line. By 13:21 UTC the observer held 280 certificates, indices 241 to 522 (DAA 7,229 to 17,982), 17 to 27 voters, no index with two hashes. Test networks, 4 October 2026, Apple M5 Max: a 4/2 split locked on the 4 side (67.9%) 2 to 8 s after the cut and never on the 2 side, 0 conflicts; a 3/3 split locked on neither side for 150 s with 0 conflicts, where the 3 October floor (56.7%) would have locked both sides at 76 and 106 s; the rule guarantees one lock history for partitions shorter than the window bound W / (3R) (200 s on that test network's 1,800-DAA window, about 40 minutes on the devnet, about 10 days at the 30-day mainnet window); beyond that bound each side can reach two thirds of its own window, so the next finality rule freezes the weight table at the last certified checkpoint and pauses instead. Simulator with the 2/3 floor: 0 conflicts up to a 33% equivocator (34% splits a 50/50 partition), silent weight pauses locks from 34%, a 50/50 partition locks alone from day 10.1. Harness: equivocating keys stripped on every node, Sybil dust at zero weight, a pulsed miner's weight equal to its block share (ratio 0.96 to 1.0), the first-month gate stops a young window locking under one key. Not demonstrated: certificate injection on the wire, an eclipse with a private fork, the 2-hour presence window at mainnet length | none yet | | 11 | Hashrate that arrived today has almost no vote: ten days of the whole network's hashrate to reach a third of the weight, twenty for two thirds; 51% never reaches two thirds while honest miners stay | Litepaper Finality; homepage firsts | tested by the team | repo `bbb264a`; `sim/finality_v2.py` | Scenario B of `sim/finality_v2.py`, seeds 7 and 11 | share(t) = (t/30) x a/(1+a) holds to 0.04 points; a renter equal to the whole honest network (a = 1) crosses 1/3 on day 20 and never reaches 2/3; a = 9 crosses 1/3 on day 11.1 and 2/3 on day 22.2. The ten-day figure is a = infinity, honest miners gone. 3 October 2026, Apple M5 Max. A model with 1,000 Pareto keys and no DAG; the live devnet's window is two hours old, so the claim has no live measurement yet | none yet | -| 12 | The difficulty rule recovers from a hashrate step within minutes, where Kaspa's sampled rule never settles. A step inside an epoch set the rule oscillating on the live devnet on 4 October 2026; rule v2 removes it in the simulator and on a test network and is built but not yet rolled out | Spec 2.3; litepaper Speed (implied); bench page | tested by the team | repo `e9328c6`, `abb5a5d` (attacks), `67bf226` (rule v2); fork `difficulty` branch (timestamp fix) and `devnet-v4` `a21ff239` (`difficulty_v2_activation_daa`, `REF_WINDOW_V2 = 600`); `sim/difficulty/sim.py --live` | The live record `sim/difficulty/records/live-2026-10-04.csv` (8,090 headers, `pull_live.py`) and the hash-rate record beside it; `sim/difficulty/sim.py` on the synthetic set and the DAG replay; `sim/difficulty/attacks/attacks.py`; `sim/difficulty/testnet_v2.py` (3 nodes, activation at DAA 900); `cargo test --release -p kaspa-consensus --lib difficulty` (15 pass); bench-log "difficulty controller", "difficulty rule under attack", "timestamp attack fixed", "difficulty rule v2" | Live devnet v4, 4 October 2026 (UTC): a second RTX 5090 joining 7 minutes into an epoch (about 152 to 280 MH/s) hardened the difficulty 70M to 144M in 90 s and then swung by about a third for 40 minutes around the true level of 139M while the epoch-long reference lane carried the join; that card leaving for 4 minutes eased 116M to 67M and back to 106M; the epoch boundary with both PCs restarting took 152M to 77M in 3 minutes, after which the rule held within 1.3% per minute with no flips. Cause: the reference lane covered the whole epoch, so a mid-epoch step polluted it for the hour and the 25% trigger flipped on the short lane's noise. The DAG replay reproduces the record (std of log difficulty 0.115 against 0.134, 4.3 peaks against 4). Rule v2 (reference window 600 DAA) on the replay: std 0.026, 0 flips, mean 142.6M against 139M true; on a 3-node test network the v2 nodes eased a leave with no peak and held a rejoin within 3% after 60 s, and a node without the activation height forked off at it as designed. Rule v2 rolled onto the 12-node cloud network on 4 October (all nodes crossed the height on one chain; a hash-rate step then settled in 160 to 270 s with no swing) and activates on the devnet at DAA 33,000 the same evening. Timestamp forging (ledger M23) fixed the same day: a 50% forger drifts the rate under 1.1% where the 3 October rule gave it a 9.9x difficulty. Simulator, settled seconds: x50 step 62 to 66 (Kaspa 1,542), /50 step 657 to 753 (Kaspa 12,296). Apple M5 Max under load 7 to 442; the DAG model is fitted on one scale; the pool hopper's 0.7-point excess over Kaspa's rule stays open | none yet | +| 12 | The difficulty rule recovers from a hashrate step within minutes, where Kaspa's sampled rule never settles. A step inside an epoch set the rule oscillating on the live devnet on 4 October 2026; rule v2 removes it in the simulator and on a test network and is built but not yet rolled out | Spec 2.3; litepaper Speed (implied); bench page | tested by the team | repo `e9328c6`, `abb5a5d` (attacks), `67bf226` (rule v2); fork `difficulty` branch (timestamp fix) and `devnet-v4` `a21ff239` (`difficulty_v2_activation_daa`, `REF_WINDOW_V2 = 600`); `sim/difficulty/sim.py --live` | The live record `sim/difficulty/records/live-2026-10-04.csv` (8,090 headers, `pull_live.py`) and the hash-rate record beside it; `sim/difficulty/sim.py` on the synthetic set and the DAG replay; `sim/difficulty/attacks/attacks.py`; the difficulty simulator's three-node v2 run (`sim/difficulty/`, activation at DAA 900); `cargo test --release -p kaspa-consensus --lib difficulty` (15 pass); bench-log "difficulty controller", "difficulty rule under attack", "timestamp attack fixed", "difficulty rule v2" | Live devnet v4, 4 October 2026 (UTC): a second RTX 5090 joining 7 minutes into an epoch (about 152 to 280 MH/s) hardened the difficulty 70M to 144M in 90 s and then swung by about a third for 40 minutes around the true level of 139M while the epoch-long reference lane carried the join; that card leaving for 4 minutes eased 116M to 67M and back to 106M; the epoch boundary with both PCs restarting took 152M to 77M in 3 minutes, after which the rule held within 1.3% per minute with no flips. Cause: the reference lane covered the whole epoch, so a mid-epoch step polluted it for the hour and the 25% trigger flipped on the short lane's noise. The DAG replay reproduces the record (std of log difficulty 0.115 against 0.134, 4.3 peaks against 4). Rule v2 (reference window 600 DAA) on the replay: std 0.026, 0 flips, mean 142.6M against 139M true; on a 3-node test network the v2 nodes eased a leave with no peak and held a rejoin within 3% after 60 s, and a node without the activation height forked off at it as designed. Rule v2 rolled onto the 12-node cloud network on 4 October (all nodes crossed the height on one chain; a hash-rate step then settled in 160 to 270 s with no swing) and activates on the devnet at DAA 33,000 the same evening. Timestamp forging (ledger M23) fixed the same day: a 50% forger drifts the rate under 1.1% where the 3 October rule gave it a 9.9x difficulty. Simulator, settled seconds: x50 step 62 to 66 (Kaspa 1,542), /50 step 657 to 753 (Kaspa 12,296). Apple M5 Max under load 7 to 442; the DAG model is fitted on one scale; the pool hopper's 0.7-point excess over Kaspa's rule stays open | none yet | | 13 | Every node executes the ordered transactions natively and reaches the same state root | Litepaper Proving ("Every node executes ... natively"), Building ("runs on Igneum unchanged") | tested by the team | repo `f5f8c80`, `8dae48b`; fork `devnet-v4` `dc749905`; revm 43.0.3 | `node tools/evm-smoke/smoke.mjs` against a 3-node `igneumd`; `igneum-exec-diff seq.json`; bench-log "execution layer devnet v3" and "devnet-v4 integration" | Simnet, 3 October 2026: 87 of 87 viem checks, state roots identical on 3 nodes at four heights, 57 executed and 19 skipped transactions agree with plain revm, 0 mismatches. Merged node on real proof of work, 4 October 2026: 84 of 85 checks (the miss needs parallel blocks the network did not produce in 36 s), 59 transfers in 10 chain blocks, state roots identical on 3 nodes, `igneum-exec-diff` 0 mismatches over 59 transactions; the live devnet v4 runs this execution layer. Apple M5 Max. The prover is a stub; state is rebuilt from genesis at start; no EVM transaction relay between nodes | none yet | | 14 | Ethereum bytecode runs unchanged, with the documented differences of spec 7.1 | Homepage Build card; litepaper Building | tested by the team | as row 13; fixes `F-exec-A`, `F-exec-B` (spec 7.5) | `tools/evm-smoke/smoke.mjs`: deploy via viem, `increment`, `hashLoop`, `eth_estimateGas`, `eth_getLogs`; `tools/exec-attacks` scenarios 1 and 3; bench-log "execution layer attack fixes" | Deployment, calls, reverts, logs and gas estimates behave as viem expects; chain id 4463 at that version (4464 since the class v5 floor, 8 October 2026); the prototype pgas table gives 0.0095 to 0.028 pgas per gas, below the design's band before calibration, 3 October 2026. 4 October 2026: a transaction that would cross the block's proving budget is refused by the mempool and, if forced in, aborted and charged with its nonce advanced (25 of 25 checks; 30 of 30 malformed cases). Apple M5 Max. The `Prover` precompile, proof records and the shard planner are not in the node | none yet | | 15 | Every block is proven, with the proof landing within about a minute at launch | Homepage stats ("~60 s to a proof"); litepaper Proving; roadmap phase 3 gate | implemented | repo `d7e1f89` (GPU proof), `e01a3cc`, `292e800`, `eedd136` (`proving/igneum-prove`: shard cutter, MPT witnesses, shard and aggregator guests); SP1 6.8.1; spec 7.2, 7.6 | `proving/windows-wsl2` (SETUP-PROVER, PROVE-BLOCK) on the RTX 5090; `igneum-prove-host --mode block` on `proving/fixtures/`; bench-log "proving v0 on the RTX 5090" and "proving: devnet v4 shards" | First GPU proof of an Igneum block, 4 October 2026, RTX 5090 (WSL2, SP1 cuda, mining paused): fixture `block-78-increment` (2 transactions), core proof 1.4 s (7.3 MB, verify 0.221 s), compressed proof 2.7 s (1.27 MB, verify 0.038 s), post-state and receipts roots identical to the node's; 15.7x and 20.6x faster than a loaded M5 Max CPU. The same day on that CPU (load 38 to 47): a three-shard block proved shard by shard and aggregated by recursion, 19 min (1,139 s) end to end, 245 to 337 s per compressed shard proof, every proof verified. What is not there: no proof is produced, carried or checked on the chain (the devnet prover is a stub that signs claims), the proving pool pays nobody (row 21), the block proven is far below one shard, and the 60-second figure remains a design target; the pass mark is the standard in `docs/benchmarks/proving-e2e.md`. Second RTX 5090 run, 4 October 2026 evening (job run-20261004-173115): a full shard at the provisional S_p (6.75 M pgas, 60.8 M cycles) executed in 1.63 s, core proof 8.3 s (18.1 MB), compressed proof 10.9 s (1.27 MB, verify 0.040 s); a two-shard block (13.5 M pgas) proved shard by shard (11.7 s and 10.0 s) and aggregated in 2.2 s, 24 s of GPU stages end to end, every proof verified, six tampered witnesses rejected. The two host defects (an abort after the upload, an idle wait that turned out to be an unbuffered 18 MB proof save through the WSL2 file bridge, 24 minutes) are fixed (ledger P20) 5 October 2026, live devnet with real transactions (bench-log "real transactions, the first non-empty shard proven and paid"): block 72704 shard 0, 29 transfers, 5,800 pgas, proven on PC 2 in 34 s, verified on the Mac in 0.297 s and paid 1.7623 IGN, 53 s after the chain block executed; of about 1,400 blocks in the 20-minute window 36 were proven (the one prover takes the newest shard assigned to it), so "every block" is not yet true; a second content shard (72803, all copies skipped) failed the native-execution veto on the exporter's block structure, fixed with fixtures the same day, the node side pending the 0.3.9 rollout 5 October 2026, evening (bench-log "proving v1"): the aggregated segment record, the chain rule and the unproven rule are implemented behind `proving_v1_activation_daa` (branch proving-v1, not on the devnet before 0.3.11); on the RTX 5090 a chain of 8 consecutive live blocks proved and aggregated by recursion in 135.6 s with the miner on the card (17 s a block, one proof of 1,272,909 bytes attesting all 8, verified in 0.04 s); the 3-node fast-time harness paid a segment record 1.0 s after submission and refused a late one after its deadline (21 checks); the devnet itself, with one prover, carried proofs for 2.4% of blocks over 30 minutes at a block-to-record latency p50 44 s, p99 52 s. The "within about a minute" holds per proven block; "every block" needs 18 mining 5090s or 6 proving-only cards at empty blocks on the measured rates, and the mandatory rule stays off until the share is one | none yet | | 16 | A 12 GB card proves one shard in about 20 s (WITHDRAWN 5 October 2026: a 24 GB card proves a full shard at the adopted size in 4.3 s; 32 GB mines and proves) | Litepaper Proving ("The proving budget"); roadmap gate 2 | designed | spec 5.1 (Target), 7.6 (`S_p` provisional, 7,500,000 pgas = `B_p` / 4) | `PROVE-SHARD.bat` on the RTX 5090 (pending); the end-to-end standard in `docs/benchmarks/proving-e2e.md`; bench-log "proving: devnet v4 shards" | Measured on a 32 GB card, not yet on a 12 GB card. A shard at the provisional `S_p` is 60.8 M SP1 cycles on the prototype pgas table (9 cycles per pgas, 44 per EVM gas; the modexp entry about 100x its SP1 cost); on an RTX 5090 (4 October 2026 evening, job run-20261004-173115) it executed in 1.63 s and its compressed proof took 10.9 s, verified in 0.040 s, so the 32 GB card is inside the 20 s target with margin. Whether a 12 GB card proves it at all, and in what time, is the next measurement (an RTX 3060 and an RTX 5060 Ti 16 GB are on order). A per-shard time can be met by shrinking the shard, so the project does not use it as a pass mark 5 October 2026, evening (bench-log "proving v1", the S_p curve): measured on the RTX 5090 with SP1 6.8.1's GPU prover, the card to itself, 1-s nvidia-smi samples: an empty shard 13,874 MiB and 2.2 s; a full shard at the ADOPTED v1 budget (30,000 pgas, 4.7 M cycles) 20,434 MiB and 4.3 s; the full prototype shard (6.75 M pgas, 60 M cycles) 28,307 MiB and 10.8 s; beside the miner 15,670 and 30,039 MiB. No environment knob of SP1 moves the 13.9 GB floor and the GPU server has no options of its own, so on this build a 12 GB card proves nothing, a 16 GB card only empty shards, a 24 GB card the adopted full shard alone and beside the miner (22,210 MiB and 13.2 s, measured on the 32 GB card: the 5090's allocation pattern, not yet a run on a 24 GB card) and a 32 GB card the prototype shard beside the miner with 2.5 GB spare. The litepaper line now says so; the 12 GB gate returns when a prover build with a smaller floor is measured on a 12 GB card | none yet | -| 17 | The chip resistance claim, served as the class v6 close words it, the claim statement above the sentence: Igneum remains competitive on accessible commodity GPUs even when specialised mining hardware is assumed to exist, remain compatible and seek profit; its security does not rely on identifying that hardware or retiring it through emergency changes. Class v6 adopts the 64-register window and retains it across every rotation. Current modelling places the strongest specialised designs assessed against the GPU tier at about 2.3x to 3.3x energy-efficiency advantage a node ahead (2.0x to 2.9x on the GPU's own node), a bracket that is approximate and provisional until the placed gated core rows land. The long-program and select-tree proposals were rejected. Economic resistance depends on development cost, deployment economics and productive hardware lifetime; family transitions receive an obsolescence benefit only where a loss of competitiveness is demonstrated; programmable multi-epoch designs are included in the assessment. Beside it: class v4 is live from the first block on the testnet and the mainnet; class v5 makes the dataset the chain's state so a stateless or stale chip is wrong on every item; the hot-set cache is bounded at 1.067x at the ceiling and the weak-day FPGA at 12 percent on 15 days a century, both routed to the next class; datacentre silicon does not change the question; the three statements (energy resistance, economic resistance, response capability) are served separate, with the harness and the scoring rule linked | the home page's chip line, the litepaper's chip section (/litepaper#chip-model), the miner page's line | the GPU side tested by the team (the RTX 5080 and RTX 5090 clock-lock passes under class v4, every card, the verifier, the two attack-pass bounds, the H100); the chip core synthesised on ASAP7 and scaled to N3, claimed; the chip's memory modelled; the node column claimed scaling; the economic surface modelled, first cut, conditional; the rotation schedule measured per boundary; class v5 designed; the Antminer X5 an observed comparison, not a ceiling; the X9 a withdrawn pre-order, never benchmarked | `docs/design/class-v6-rotating-family.md` section 10 (10.0 to 10.0h, the close and its two accepted external reviews, 8 October 2026); `docs/design/class-v5-stored-state.md` sections 0, 13 and 14 and `docs/design/class-v5-harness/` (branch class-v5); `docs/design/class-rotation-four-layers.md`; `docs/analysis/chip-model-v3.md` 5 and 6; `docs/analysis/latency-shadow-2026-10-06.md`; `docs/analysis/attack-pass/f8-uniform.md`, `f4-weakday.md`, `docs/analysis/ca3-v4-uniform.md`; the H100 row of 7 October; `docs/plans/cryptanalysis/in-house-pass.md` (the internal adversarial pass) | the scoring rule in the close (the minimum over workloads of the maximum over free adversarial designs of the GPU's joules per hash over the adversary's, under the 10 percent GPU-cost budget at the lock, the verifier limit, cross-vendor correctness and hardware accessibility); the card rows by the benchmark package; the class v5 harness and the family harness; the attack-pass harnesses `tools/attack/f8-uniform` and the F4 census; the verifier by `igneum-pow bench` | 136 MH/s at 350 W (5090, bench) and 290 W (app); the class v4 efficiency passes (bench log "7 to 8 October 2026, the class v4 efficiency passes: the core clock lock on the RTX 5090 and the RTX 5080", measured): the 5090 at 136.84 MH/s and 475.5 W unlocked, 134.98 at 316.3 W at a 1,400 MHz core lock, the best points class v4 at 1,200 MHz (133.80 MH/s, 305.1 W, 0.439 MH/W) and class v3 at 1,300 MHz (134.62, 223.3 W, 0.603), the premium 145 W unlocked and 82 W at the best points, the knee 1,300 MHz; the RTX 5080 (8 October 2026, the dock card of the three-card Windows rig) at 71.41 MH/s and 253.1 W unlocked under class v4 against 71.28 at 169.7 W under class v3, the best points class v4 at 1,100 MHz (71.20 MH/s, 146.6 W, 0.486 MH/W) and class v3 at 1,000 MHz (71.11, 103.7 W, 0.686), the premium 83.4 W unlocked and 41 W at the best points, the knee between 1,000 and 900 MHz; per tier: a 5080 owner on class v4 locked near 1,100 MHz pays 147 W instead of 253 for 0.3 percent less rate, MH per watt up 72 percent, the lever Ember Tune's core-clock knob in 0.3.24; 27 MH/s at 21 W (M5 Max); 249 MH/s (H100 SXM) at 98 percent of its read ceiling, 1.78x hash, 1.15x MH/W, a third per rented dollar; 2.33 ms per warp; the GPU side of the close: the RTX 5080 at its 1,100 MHz lock 2.06 microjoules per hash and the RTX 5090 at its 1,300 MHz lock 2.33 (class v4, 8 October 2026); the modelled bracket about 2.3x to 3.3x a node ahead and 2.0x to 2.9x node for node (approximate, provisional; the clock-gated base core k about 0.37 at N3, the gated window adding about 0.13; the first placed core 64 percent above synthesis), the placed gated row expected near 2.6x to 3.1x and 2.3x to 2.7x and served when it lands; the shadow's premium on a 5090 81.8 W at its knee (class v4 at the 1,200 MHz lock 133.80 MH/s at 305.1 W; class v3 at 1,300 MHz 134.62 at 223.3 W; 7 October 2026); 1.067x at the ceiling; 12 percent on 15 days a century; 10.85 ms at rung 3; 6 to 8 October 2026, the M5 Max, the desk rigs' RTX 5090, RTX 5080, RX 9070 XT and RTX 4070, rented pods, a rented H100 SXM, igneum-build-1 Bitmain's Antminer X9 (RandomX; 1,000 KH/s, 2,472 W, 2.47 J per KH, USD 5,600; pre-orders 26 December 2025) was withdrawn in mid-May 2026 with buyers refunded before any unit shipped, no independent benchmark, commodity Sophgo SG2044 server SoCs with an AES accelerator, no tapeout: it is a precedent on the served pages, not a chip core against the model (attack pass AP-F5-1, 7 October 2026; the close's 10.0f, 8 October 2026). Withdrawn from the served pages on 8 October 2026 and not restated: the 2.1x and 3.4x launch line, the 5x to 9x class v3 baseline, the ladder's 2.8x row, the USD 100 M pay-back row, any chip-arrival probability, the lifetime claim and the USD 300 M and 340 M lines. | none yet; the next test is the internal adversarial pass (three lanes new to the hash code, outsider inputs only, reports published whole), whose floor reading is in (measured, 8 October 2026; ledger AP-F8-1 and AP-F8-6): nine of nine hot sets refused; the diffuse era-stride excess, bounded under 0.1 percent of a hash's reads per site, is not caught by the floor and is the next class's test, and no outside review has run yet | +| 17 | The chip resistance claim, served as the class v6 close words it, the claim statement above the sentence: Igneum remains competitive on accessible commodity GPUs even when specialised mining hardware is assumed to exist, remain compatible and seek profit; its security does not rely on identifying that hardware or retiring it through emergency changes. Class v6 adopts the 64-register window and retains it across every rotation. Current modelling places the strongest specialised designs assessed against the GPU tier at about 2.3x to 3.3x energy-efficiency advantage a node ahead (2.0x to 2.9x on the GPU's own node), a bracket that is approximate and provisional until the placed gated core rows land. The long-program and select-tree proposals were rejected. Economic resistance depends on development cost, deployment economics and productive hardware lifetime; family transitions receive an obsolescence benefit only where a loss of competitiveness is demonstrated; programmable multi-epoch designs are included in the assessment. Beside it: class v5 derives the dataset from chain state; whether that excludes a specialised design is under evaluation (Deliverable 3), since a design that tracks state is not excluded by staleness; the hot-set cache is bounded at 1.067x at the ceiling and the weak-day FPGA at 12 percent on 15 days a century, both routed to the next class; datacentre silicon does not change the question; the three statements (energy resistance, economic resistance, response capability) are served separate, with the harness and the scoring rule linked; the energy ratio is not the pass criterion: the coexistence model ([docs/analysis/class-v6/coexistence-model.md](https://git.igneum.network/igneum-network/igneum/src/branch/master/docs/analysis/class-v6/coexistence-model.md)) is, and its first run's result is served with its conditions; the window sentence and the bracket stay provisional until the placed gated rows land | the home page's chip line, the litepaper's chip section (/litepaper#chip-model), the miner page's line | the GPU side tested by the team (the RTX 5080 and RTX 5090 clock-lock passes under class v4, every card, the verifier, the two attack-pass bounds, the H100); the chip core synthesised on ASAP7 and scaled to N3, claimed; the chip's memory modelled; the node column claimed scaling; the economic surface modelled, first cut, conditional; the rotation schedule measured per boundary; class v5 designed; the Antminer X5 an observed comparison, not a ceiling; the X9 a withdrawn pre-order, never benchmarked | `docs/design/class-v6-rotating-family.md` section 10 (10.0 to 10.0h, the close and its two accepted external reviews, 8 October 2026); `docs/design/class-v5-stored-state.md` sections 0, 13 and 14 and `docs/design/class-v5-harness/` (branch class-v5); `docs/design/class-rotation-four-layers.md`; `docs/analysis/chip-model-v3.md` 5 and 6; `docs/analysis/latency-shadow-2026-10-06.md`; `docs/analysis/attack-pass/f8-uniform.md`, `f4-weakday.md`, `docs/analysis/ca3-v4-uniform.md`; the H100 row of 7 October; `docs/plans/cryptanalysis/in-house-pass.md` (the internal adversarial pass) | the scoring rule in the close (the minimum over workloads of the maximum over free adversarial designs of the GPU's joules per hash over the adversary's, under the 10 percent GPU-cost budget at the lock, the verifier limit, cross-vendor correctness and hardware accessibility); the card rows by the benchmark package; the class v5 harness and the family harness; the attack-pass harnesses `tools/attack/f8-uniform` and the F4 census; the verifier by `igneum-pow bench` | 136 MH/s at 350 W (5090, bench) and 290 W (app); the class v4 efficiency passes (bench log "7 to 8 October 2026, the class v4 efficiency passes: the core clock lock on the RTX 5090 and the RTX 5080", measured): the 5090 at 136.84 MH/s and 475.5 W unlocked, 134.98 at 316.3 W at a 1,400 MHz core lock, the best points class v4 at 1,200 MHz (133.80 MH/s, 305.1 W, 0.439 MH/W) and class v3 at 1,300 MHz (134.62, 223.3 W, 0.603), the premium 145 W unlocked and 82 W at the best points, the knee 1,300 MHz; the RTX 5080 (8 October 2026, the dock card of the three-card Windows rig) at 71.41 MH/s and 253.1 W unlocked under class v4 against 71.28 at 169.7 W under class v3, the best points class v4 at 1,100 MHz (71.20 MH/s, 146.6 W, 0.486 MH/W) and class v3 at 1,000 MHz (71.11, 103.7 W, 0.686), the premium 83.4 W unlocked and 41 W at the best points, the knee between 1,000 and 900 MHz; per tier: a 5080 owner on class v4 locked near 1,100 MHz pays 147 W instead of 253 for 0.3 percent less rate, MH per watt up 72 percent, the lever Ember Tune's core-clock knob in 0.3.24; 27 MH/s at 21 W (M5 Max); 249 MH/s (H100 SXM) at 98 percent of its read ceiling, 1.78x hash, 1.15x MH/W, a third per rented dollar; 2.33 ms per warp; the GPU side of the close: the RTX 5080 at its 1,100 MHz lock 2.06 microjoules per hash and the RTX 5090 at its 1,300 MHz lock 2.33 (class v4, 8 October 2026); the modelled bracket about 2.3x to 3.3x a node ahead and 2.0x to 2.9x node for node (approximate, provisional; the clock-gated base core k about 0.37 at N3, the gated window adding about 0.13; the first placed core 64 percent above synthesis), the placed gated row expected near 2.6x to 3.1x and 2.3x to 2.7x and served when it lands; the shadow's premium on a 5090 81.8 W at its knee (class v4 at the 1,200 MHz lock 133.80 MH/s at 305.1 W; class v3 at 1,300 MHz 134.62 at 223.3 W; 7 October 2026); 1.067x at the ceiling; 12 percent on 15 days a century; 10.85 ms at rung 3; 6 to 8 October 2026, the M5 Max, the desk rigs' RTX 5090, RTX 5080, RX 9070 XT and RTX 4070, rented pods, a rented H100 SXM, igneum-build-1 Bitmain's Antminer X9 (RandomX; 1,000 KH/s, 2,472 W, 2.47 J per KH, USD 5,600; pre-orders 26 December 2025) was withdrawn in mid-May 2026 with buyers refunded before any unit shipped, no independent benchmark, commodity Sophgo SG2044 server SoCs with an AES accelerator, no tapeout: it is a precedent on the served pages, not a chip core against the model (attack pass AP-F5-1, 7 October 2026; the close's 10.0f, 8 October 2026). Withdrawn from the served pages on 8 October 2026 and not restated: the 2.1x and 3.4x launch line, the 5x to 9x class v3 baseline, the ladder's 2.8x row, the USD 100 M pay-back row, any chip-arrival probability, the lifetime claim and the USD 300 M and 340 M lines. | none yet; the next test is the internal adversarial pass (three lanes new to the hash code, outsider inputs only, reports published whole), whose floor reading is in (measured, 8 October 2026; ledger AP-F8-1 and AP-F8-6): nine of nine hot sets refused; the diffuse era-stride excess, bounded under 0.1 percent of a hash's reads per site, is not caught by the floor and is the next class's test, and no outside review has run yet | | 18 | The chip resistance measurements: the program is latency-bound (dependent reads spread over the whole dataset), not bandwidth-bound, on every card we own, and sits beyond a card's on-chip cache; measured 8 October 2026: the distinct-index floor holds at 0.995 on every accepted program, about half of epochs carry one load site with a biased address bit at the era's stride rotation, priced at about 1.6 percent of a hash's reads to a chip storing half the dataset and nothing to one storing all of it (`docs/analysis/class-v6/family-gate.md`, lane D; adv-cache-2's `report-chained-cache-2.md` section 2.3 on its branch; ledger AP-F8-7) | Litepaper Mining ("waits on memory latency, not on maths or bandwidth"), vs RandomX; the numbers page | tested by the team | readwidth e752fc7 (`docs/plans/read-width.md`), ca2-era 78c0ee4, ca2-cache 2de19e5 (`docs/plans/hot-table.md`) | The dependent-read probes at 32 to 1,024 MiB and the hash rate per class on the three cards; the latency-bound share = rate over the probe ceiling per load | Latency-bound share at the 1 GiB dataset: RTX 5090 0.96 (v2) and 1.01 (v3), RX 9070 XT 0.87 and 0.95, M5 Max 1.01 and 1.06; wider reads do not close the AMD gap (the 9070 XT does 2.4 G dependent reads per second at every width; the 5090 goes bandwidth-bound at 64 B, share 0.58); a 32 to 96 MiB hot table is not kept resident by any card while the dataset streams (g 0.80 to 0.87 in the added form). 5 October 2026 | none yet | | 19 | The lottery hash is sound as a hash: uniform output, deterministic, no out-of-bounds read, fuzzed; class v3 bit-exact on the three vendors | Litepaper vs RandomX ("Every number above is measured and logged"), the numbers page | tested by the team | ca2-mixer 1ab8b21 (`tests/mixer.rs`, `tests/scratch.rs`), ca2-era 78c0ee4, ca2-soundness a465881 (`docs/analysis/scratch-soundness.md`), `igneum-pow/tests/packs.rs` | The crate suite (53 + 4 + 19 + 7), the Metal fuzz, edge, stats and determinism runs on the v3 construction, the pack vectors and 2^24 fingerprints on Metal, Apple OpenCL, the RTX 5090 and the RX 9070 XT, the 1,024-hash CPU re-check per card | Class v3 (mixer x8 + era): 200-program fuzz 200 of 200 on Metal, every tenth on Apple OpenCL; the pinned v3 packs 3/3 + 3/3 and 96 of 96 lanes on Metal and Apple OpenCL; the six era packs' fingerprints equal on the three vendors (PC 1 job run-ca2-era-pc1-20261005, 5 October 2026); the v2 exports byte-identical on the v3 crate; the final-class PC rows and the G2 re-check: job run-ca2-era-pc1b-20261005 (pending at the time of writing) | none yet | | 20 | No premine, no pre-sale, no allocation: every coin is minted by the schedule and every coin goes to the block producer (80%) and the proving pool (20%) | Homepage stats and Economics tiles; litepaper Supply, Economics | implemented | repo `6ac80a3`; fork "igneum-node devnet v0"; `consensus/core/src/igneum.rs`, `coinbase.rs` | `cargo test -p kaspa-consensus-core igneum` (8 pass: subsidy table, ramp, split, cap) and `cargo test -p kaspa-consensus coinbase` (8 pass); `igneum-miner inspect 40`; bench-log "igneum-node devnet v0" | Coinbases on the devnet: 80/20 exact on 39 of 39 single-payee blocks, the 20% to the `igneum-proving-pool-v0` output; the per-second schedule sums to under the 4,000,000,000 cap by less than 100 coins; 3,168,808,781 units per DAA second in years 0 to 2, halving at 63,115,200 DAA s. 3 October 2026, Apple M5 Max. The devnet genesis carries no allocation; the mainnet genesis does not exist yet, so the claim is about the code and the stated rule, not a launch that has happened | none yet | @@ -95,10 +95,9 @@ Versions in the table: `igneum-pow` is the Rust crate at `igneum-pow/Cargo.toml` | Row | Before | After | Why | |---|---|---|---| -| 7 | the first devnet's block rate | Devnet 3's first block and its first minutes added | Devnet 3 started at 17:06 UTC with every upgrade on from block zero | -| 10 | rule v2, first lock 4 October | Devnet 3's first lock under rule v3 at 19:02 UTC added | the first lock on Devnet 3 | +| 7 | the first devnet's block rate | the devnet's first block and its first minutes added | the devnet started at 17:06 UTC with every upgrade on from block zero | +| 10 | rule v2, first lock 4 October | the devnet's first lock under rule v3 at 19:02 UTC added | the first lock on the devnet | | 17 | 136 MH/s at 350 W (class v3) | the class v4 efficiency pass added: 136.84 MH/s at 475.5 W unlocked, 134.98 MH/s at 316.3 W locked, control 134.68 MH/s at 228.0 W | the 5090 efficiency pass | -| rule 1 | "the repository is private until the public testnet" | the ceiling is stated without the repository's state | the repository links moved to git.igneum.network | ## What moved on 8 October 2026 diff --git a/site/claims.html b/site/claims.html index 2671e905c..509a43aee 100644 --- a/site/claims.html +++ b/site/claims.html @@ -247,7 +247,7 @@

Here are the limits, stated before anyone else states them.

Everything in this document is subject to the gates on the roadmap. Nothing in it is an offer to sell anything. Found an error, or a criticism this document does not answer? Email hello@igneum.network, or open an issue on the public specification repository: git.igneum.network/igneum-network/spec/issues. Post reaches Igneum Labs LTD, Unit IH-00-01-01-OF-01, Level 01, Innovation One, Dubai International Financial Centre.

diff --git a/site/evidence.html b/site/evidence.html index 33e6d789f..72bbed3e9 100644 --- a/site/evidence.html +++ b/site/evidence.html @@ -270,17 +270,17 @@ td.mono{font-family:var(--f-mono);font-size:12.5px;min-width:180px}td.iv{color:v 4The same program produces identical hashes on three GPU vendors, cache and dataset included
Litepaper vs RandomX ("Bit-exact on Apple, NVIDIA and AMD, measured"), For miners; homepage
tested by the teamrepo f2e903e, 0f1fdaf (version 1 packs), b27da39 (version 2 packs igneum-genesis-mh, igneum-devnet-v4-epoch0); igneum-pow 0.2.0The 96 test vectors of a pack through proto-metal/igneum-bench, proto-cuda/host.cu, proto-opencl/host.c; batch fingerprint at --batch-log2 24; the miner's CPU re-check of every share a GPU worker finds on the devnet; bench-log entries "RTX 5090, memory-hard dataset", "AMD gfx1036", "RTX 5090 through NVIDIA OpenCL", "generator version 2 adopted", "the gfx1036 worker fault"Version 1: 96/96 on Apple Metal (M5 Max), NVIDIA CUDA and NVIDIA OpenCL (RTX 5090, Windows), AMD OpenCL (Ryzen 7 9800X3D integrated gfx1036, 1 compute unit), Apple OpenCL, pocl and two CPU references; batch fingerprint 98af644e993239e2 over 16.7 million nonces identical on the AMD chip and the 5090, 3 October 2026. Version 2: 96/96 on Apple Metal, Apple OpenCL and the CUDA and OpenCL emulators with identical fingerprints; on real NVIDIA and AMD silicon the version 2 vectors have not run as a pack, but both mined accepted blocks on the live devnet with the CPU re-check clean on every share (RTX 5090 at 124.2 MH/s, gfx1036 at 3.3 MH/s), 4 October 2026. The AMD device is an integrated chip; no discrete AMD card and no Intel card has run anything (O-1.15)none yet 5A CPU verifies one hash in under 10 ms by simulating one warp
Litepaper Mining ("about ten milliseconds"), vs RandomX; roadmap gate 2
tested by the teamrepo 75cac18, b27da39; igneum-pow 0.2.0 (verify.rs)cargo test and the crate bench in igneum-pow/; bench-log "igneum-pow: Rust crate bit-exact with proto-metal" and "generator version 2 adopted"0.411 to 0.579 ms per 32-lane warp steady, 0.41 to 0.87 ms cold, average of 20, 1 GiB dataset, cache held, one M5 Max performance core, 3 October 2026; version 2 units 0.631 ms (average of 20), cold 0.67 to 0.81 ms, 4 October 2026. Gate margin about 16x on this core. Not measured on a 2019-class laptop core (O-1.14)none yet 6The hash is bound to the header: one nonce serves one header, and a wrong nonce is rejected
Spec 1.6; litepaper Mining (implied by "checks a hash")
tested by the teamrepo 33f7b33, 9812466, b27da39; igneum-pow 0.2.0 (bind.rs, bound vectors re-cut for version 2, 39 crate tests)igneum-miner bad-nonce against a devnet node; igneum-pow hash-bound for the 96-nonce job across the 2^32 lane boundary; bench-log "first devnet blocks on the real lottery hash" and "generator version 2 adopted"833 blocks accepted by igneum-lottery-v1-bound on 3 nodes, 0 rejections; bad-nonce gave Reject(BlockInvalid); Metal, OpenCL and CUDA (emulated) workers bit-exact with the crate on the lane-boundary job, 3 October 2026, Apple M5 Max. Version 2: the node's engine reports igneum-lottery-v2-bound, 39 of 39 crate tests, and the live devnet v4 accepts its blocks under it, 4 October 2026none yet -7The devnet runs at one block a second
Homepage stats ("1 / s"); litepaper Speed; roadmap phase 3
tested by the teamrepo 9812466, e9328c6, 8dae48b; fork devnet-v4 dc749905The merged node's 3-node test network (igneum-devnet-880, 960 s); the live devnet v4 record sim/difficulty/records/live-2026-10-04.csv; the 12-node cloud network's arrival logs; bench-log "devnet-v4 integration", "difficulty rule v2", "first devnet blocks"Devnet 3, 7 October 2026: first block accepted at 17:06 UTC, 85 of 85 GPU blocks by 17:10 UTC and 287 by 17:14 UTC, 0 rejected (docs/plans/release-0.3.22.md section 5). Merged node, 4 October 2026, Apple M5 Max: 1,055 blocks in 960 s, 1.03 blocks/s, sink identical on 3 nodes at 31 of 31 samples, 0 rejected. Live devnet v4 the same day: 49 to 81 blocks a minute while two RTX 5090s joined and left (row 12), 1.1 to 1.2 blocks/s in the oscillating window, then within 1.3% per minute with one PC and the Apple M5 Max. The 12-node cloud network at one block a second: 644 blocks in a 10-minute window. The 3 October CPU devnet: 1.29 blocks/s over 641 s, 1.03 after the first retarget. The phase 3 gate also asks for proofs under 60 s behind the tip; no proof is on the chain (row 15)none yet +7The devnet runs at one block a second
Homepage stats ("1 / s"); litepaper Speed; roadmap phase 3
tested by the teamrepo 9812466, e9328c6, 8dae48b; fork devnet-v4 dc749905The merged node's 3-node test network (igneum-devnet-880, 960 s); the live devnet v4 record sim/difficulty/records/live-2026-10-04.csv; the 12-node cloud network's arrival logs; bench-log "devnet-v4 integration", "difficulty rule v2", "first devnet blocks"the devnet, 7 October 2026: first block accepted at 17:06 UTC, 85 of 85 GPU blocks by 17:10 UTC and 287 by 17:14 UTC, 0 rejected (docs/plans/release-0.3.22.md section 5). Merged node, 4 October 2026, Apple M5 Max: 1,055 blocks in 960 s, 1.03 blocks/s, sink identical on 3 nodes at 31 of 31 samples, 0 rejected. Live devnet v4 the same day: 49 to 81 blocks a minute while two RTX 5090s joined and left (row 12), 1.1 to 1.2 blocks/s in the oscillating window, then within 1.3% per minute with one PC and the Apple M5 Max. The 12-node cloud network at one block a second: 644 blocks in a 10-minute window. The 3 October CPU devnet: 1.29 blocks/s over 641 s, 1.03 after the first retarget. The phase 3 gate also asks for proofs under 60 s behind the tip; no proof is on the chain (row 15)none yet 8Blocks are mined by GPUs on Apple and NVIDIA
Homepage live strip; journey phase 3 ("GPU miners on three vendors")
tested by the teamrepo 9812466, e9328c6, d7e1f89, 2309c8d; fork devnet-v4Metal worker proto-metal/igneum-bench --serve driven by igneum-miner --worker; the live devnet v4 hash-rate record sim/difficulty/records/live-2026-10-04-hashrate.csv (587 worker STATUS lines by run id); bench-log "first devnet blocks", "devnet v4 cut-over", "difficulty rule v2", "first machine on the Igneum Miner app"Metal: 506 jobs, 5,636 blocks found and accepted, 0 rejected, 0 CPU/GPU mismatches, 28.2 MH/s wall, 3 October 2026. Live devnet v4, 4 October 2026: the three-card Windows rig's RTX 5090 at 122 MH/s with 8 identities, the RTX 5090 Windows rig's at 124 MH/s with 8 identities (117 to 119 MH/s inside the one-click app, 34 accepted blocks in its first minute, CPU re-check OK on every share), the Apple M5 Max's Metal worker at 26.7 MH/s; 17 vote keys signed the first finality lock (row 10); from the afternoon an Apple silicon laptop outside the project at 21.0 MH/s through the app (row 30). Two RTX 5090s and two Apple chips; no other NVIDIA model has minednone yet 9Blocks are mined by a GPU on AMD
Journey phase 3 ("three vendors")
tested by the teamrepo 2c4b30f (generic OpenCL worker, --pack), 112acf6 (fault guards); bound kernel kernel_bound.cl in the packigneum-worker-opencl.exe --pack on the RTX 5090 Windows rig's integrated Radeon against the live devnet v4 through the Windows package; bench-log "the gfx1036 worker fault", "first hourly program swap", "first machine on the Igneum Miner app"the RTX 5090 Windows rig's integrated gfx1036 (1 compute unit) mined on the live devnet on 4 October 2026: 8 accepted blocks at 3.3 MH/s over 577 s with the CPU re-check clean, and 2.74 MH/s through the hourly program swap with 0 rejected. At about 600 s the AMD runtime began answering every call with success while running nothing (906 jobs became 56,384 in 30 s, 4.3 GH/s of phantom work); not reproduced on Apple OpenCL in 4,565 jobs with 0 leaked objects; the worker and miner now refuse a job 20x faster than the mean or an unchanged output buffer and restart (112acf6), and the next gfx1036 run names the guard that fires. One integrated chip; no discrete AMD card has run anythingnone yet -10Checkpoints lock every 30 s of chain at two thirds of all 30-day weight, and the floor stops conflicting locks in partitions and eclipses for as long as neither side's own new blocks carry it past two thirds of its window (about 10 days of a 30-day window at a 50/50 split)
Litepaper Finality, "What Igneum does not claim"; homepage "locked every 30 seconds"
tested by the teamrepo a3a9833 (2/3 floor, O-3.15), bbb264a (simulation), c16ccf1; fork devnet-v4 6457ca95 (FLOOR_NUM / FLOOR_DEN 2/3), da1eb889 (F17 by-weight sortition, F1 first-month gate min_daa = window); spec 3.3, 3.3.1, 3.7, 3.9The live devnet v4 (getFinalityCheckpoints, tools/observer/observer.mjs, /api/checkpoint); sim/finality_v2.py --floor 1.0, scenarios A to L; the three-node, six-voter partition runs igneum-devnet-921 to -923; tools/finality-attacks scenarios 1 to 6 and 8; bench-log "first finality lock on the live devnet", "finality floor 2/3", "finality v2 attack harness", "finality fixes F17 and F1"Devnet 3, 7 October 2026: finality rule v3 from block zero, first lock at 19:02 UTC, under two hours after the first block (docs/plans/release-0.3.22.md). The first devnet: the first lock on the live devnet was checkpoint 242 at 11:03:44 UTC on 4 October 2026, two hours after genesis (the window and min_daa are 7,200 DAA), with 77.4% of all weight and of active weight signed by 12 aggregated votes from 17 vote keys; observer.mjs saw it 0.7 s after the miner's own lock line. By 13:21 UTC the observer held 280 certificates, indices 241 to 522 (DAA 7,229 to 17,982), 17 to 27 voters, no index with two hashes. Test networks, 4 October 2026, Apple M5 Max: a 4/2 split locked on the 4 side (67.9%) 2 to 8 s after the cut and never on the 2 side, 0 conflicts; a 3/3 split locked on neither side for 150 s with 0 conflicts, where the 3 October floor (56.7%) would have locked both sides at 76 and 106 s; the rule guarantees one lock history for partitions shorter than the window bound W / (3R) (200 s on that test network's 1,800-DAA window, about 40 minutes on the devnet, about 10 days at the 30-day mainnet window); beyond that bound each side can reach two thirds of its own window, so the next finality rule freezes the weight table at the last certified checkpoint and pauses instead. Simulator with the 2/3 floor: 0 conflicts up to a 33% equivocator (34% splits a 50/50 partition), silent weight pauses locks from 34%, a 50/50 partition locks alone from day 10.1. Harness: equivocating keys stripped on every node, Sybil dust at zero weight, a pulsed miner's weight equal to its block share (ratio 0.96 to 1.0), the first-month gate stops a young window locking under one key. Not demonstrated: certificate injection on the wire, an eclipse with a private fork, the 2-hour presence window at mainnet lengthnone yet +10Checkpoints lock every 30 s of chain at two thirds of all 30-day weight, and the floor stops conflicting locks in partitions and eclipses for as long as neither side's own new blocks carry it past two thirds of its window (about 10 days of a 30-day window at a 50/50 split)
Litepaper Finality, "What Igneum does not claim"; homepage "locked every 30 seconds"
tested by the teamrepo a3a9833 (2/3 floor, O-3.15), bbb264a (simulation), c16ccf1; fork devnet-v4 6457ca95 (FLOOR_NUM / FLOOR_DEN 2/3), da1eb889 (F17 by-weight sortition, F1 first-month gate min_daa = window); spec 3.3, 3.3.1, 3.7, 3.9The live devnet v4 (getFinalityCheckpoints, tools/observer/observer.mjs, /api/checkpoint); sim/finality_v2.py --floor 1.0, scenarios A to L; the three-node, six-voter partition runs igneum-devnet-921 to -923; tools/finality-attacks scenarios 1 to 6 and 8; bench-log "first finality lock on the live devnet", "finality floor 2/3", "finality v2 attack harness", "finality fixes F17 and F1"the devnet, 7 October 2026: finality rule v3 from block zero, first lock at 19:02 UTC, under two hours after the first block (docs/plans/release-0.3.22.md). The first devnet: the first lock on the live devnet was checkpoint 242 at 11:03:44 UTC on 4 October 2026, two hours after genesis (the window and min_daa are 7,200 DAA), with 77.4% of all weight and of active weight signed by 12 aggregated votes from 17 vote keys; observer.mjs saw it 0.7 s after the miner's own lock line. By 13:21 UTC the observer held 280 certificates, indices 241 to 522 (DAA 7,229 to 17,982), 17 to 27 voters, no index with two hashes. Test networks, 4 October 2026, Apple M5 Max: a 4/2 split locked on the 4 side (67.9%) 2 to 8 s after the cut and never on the 2 side, 0 conflicts; a 3/3 split locked on neither side for 150 s with 0 conflicts, where the 3 October floor (56.7%) would have locked both sides at 76 and 106 s; the rule guarantees one lock history for partitions shorter than the window bound W / (3R) (200 s on that test network's 1,800-DAA window, about 40 minutes on the devnet, about 10 days at the 30-day mainnet window); beyond that bound each side can reach two thirds of its own window, so the next finality rule freezes the weight table at the last certified checkpoint and pauses instead. Simulator with the 2/3 floor: 0 conflicts up to a 33% equivocator (34% splits a 50/50 partition), silent weight pauses locks from 34%, a 50/50 partition locks alone from day 10.1. Harness: equivocating keys stripped on every node, Sybil dust at zero weight, a pulsed miner's weight equal to its block share (ratio 0.96 to 1.0), the first-month gate stops a young window locking under one key. Not demonstrated: certificate injection on the wire, an eclipse with a private fork, the 2-hour presence window at mainnet lengthnone yet 11Hashrate that arrived today has almost no vote: ten days of the whole network's hashrate to reach a third of the weight, twenty for two thirds; 51% never reaches two thirds while honest miners stay
Litepaper Finality; homepage firsts
tested by the teamrepo bbb264a; sim/finality_v2.pyScenario B of sim/finality_v2.py, seeds 7 and 11share(t) = (t/30) x a/(1+a) holds to 0.04 points; a renter equal to the whole honest network (a = 1) crosses 1/3 on day 20 and never reaches 2/3; a = 9 crosses 1/3 on day 11.1 and 2/3 on day 22.2. The ten-day figure is a = infinity, honest miners gone. 3 October 2026, Apple M5 Max. A model with 1,000 Pareto keys and no DAG; the live devnet's window is two hours old, so the claim has no live measurement yetnone yet -12The difficulty rule recovers from a hashrate step within minutes, where Kaspa's sampled rule never settles. A step inside an epoch set the rule oscillating on the live devnet on 4 October 2026; rule v2 removes it in the simulator and on a test network and is built but not yet rolled out
Spec 2.3; litepaper Speed (implied); bench page
tested by the teamrepo e9328c6, abb5a5d (attacks), 67bf226 (rule v2); fork difficulty branch (timestamp fix) and devnet-v4 a21ff239 (difficulty_v2_activation_daa, REF_WINDOW_V2 = 600); sim/difficulty/sim.py --liveThe live record sim/difficulty/records/live-2026-10-04.csv (8,090 headers, pull_live.py) and the hash-rate record beside it; sim/difficulty/sim.py on the synthetic set and the DAG replay; sim/difficulty/attacks/attacks.py; sim/difficulty/testnet_v2.py (3 nodes, activation at DAA 900); cargo test --release -p kaspa-consensus --lib difficulty (15 pass); bench-log "difficulty controller", "difficulty rule under attack", "timestamp attack fixed", "difficulty rule v2"Live devnet v4, 4 October 2026 (UTC): a second RTX 5090 joining 7 minutes into an epoch (about 152 to 280 MH/s) hardened the difficulty 70M to 144M in 90 s and then swung by about a third for 40 minutes around the true level of 139M while the epoch-long reference lane carried the join; that card leaving for 4 minutes eased 116M to 67M and back to 106M; the epoch boundary with both PCs restarting took 152M to 77M in 3 minutes, after which the rule held within 1.3% per minute with no flips. Cause: the reference lane covered the whole epoch, so a mid-epoch step polluted it for the hour and the 25% trigger flipped on the short lane's noise. The DAG replay reproduces the record (std of log difficulty 0.115 against 0.134, 4.3 peaks against 4). Rule v2 (reference window 600 DAA) on the replay: std 0.026, 0 flips, mean 142.6M against 139M true; on a 3-node test network the v2 nodes eased a leave with no peak and held a rejoin within 3% after 60 s, and a node without the activation height forked off at it as designed. Rule v2 rolled onto the 12-node cloud network on 4 October (all nodes crossed the height on one chain; a hash-rate step then settled in 160 to 270 s with no swing) and activates on the devnet at DAA 33,000 the same evening. Timestamp forging (ledger M23) fixed the same day: a 50% forger drifts the rate under 1.1% where the 3 October rule gave it a 9.9x difficulty. Simulator, settled seconds: x50 step 62 to 66 (Kaspa 1,542), /50 step 657 to 753 (Kaspa 12,296). Apple M5 Max under load 7 to 442; the DAG model is fitted on one scale; the pool hopper's 0.7-point excess over Kaspa's rule stays opennone yet +12The difficulty rule recovers from a hashrate step within minutes, where Kaspa's sampled rule never settles. A step inside an epoch set the rule oscillating on the live devnet on 4 October 2026; rule v2 removes it in the simulator and on a test network and is built but not yet rolled out
Spec 2.3; litepaper Speed (implied); bench page
tested by the teamrepo e9328c6, abb5a5d (attacks), 67bf226 (rule v2); fork difficulty branch (timestamp fix) and devnet-v4 a21ff239 (difficulty_v2_activation_daa, REF_WINDOW_V2 = 600); sim/difficulty/sim.py --liveThe live record sim/difficulty/records/live-2026-10-04.csv (8,090 headers, pull_live.py) and the hash-rate record beside it; sim/difficulty/sim.py on the synthetic set and the DAG replay; sim/difficulty/attacks/attacks.py; the difficulty simulator's three-node v2 run (sim/difficulty/, activation at DAA 900); cargo test --release -p kaspa-consensus --lib difficulty (15 pass); bench-log "difficulty controller", "difficulty rule under attack", "timestamp attack fixed", "difficulty rule v2"Live devnet v4, 4 October 2026 (UTC): a second RTX 5090 joining 7 minutes into an epoch (about 152 to 280 MH/s) hardened the difficulty 70M to 144M in 90 s and then swung by about a third for 40 minutes around the true level of 139M while the epoch-long reference lane carried the join; that card leaving for 4 minutes eased 116M to 67M and back to 106M; the epoch boundary with both PCs restarting took 152M to 77M in 3 minutes, after which the rule held within 1.3% per minute with no flips. Cause: the reference lane covered the whole epoch, so a mid-epoch step polluted it for the hour and the 25% trigger flipped on the short lane's noise. The DAG replay reproduces the record (std of log difficulty 0.115 against 0.134, 4.3 peaks against 4). Rule v2 (reference window 600 DAA) on the replay: std 0.026, 0 flips, mean 142.6M against 139M true; on a 3-node test network the v2 nodes eased a leave with no peak and held a rejoin within 3% after 60 s, and a node without the activation height forked off at it as designed. Rule v2 rolled onto the 12-node cloud network on 4 October (all nodes crossed the height on one chain; a hash-rate step then settled in 160 to 270 s with no swing) and activates on the devnet at DAA 33,000 the same evening. Timestamp forging (ledger M23) fixed the same day: a 50% forger drifts the rate under 1.1% where the 3 October rule gave it a 9.9x difficulty. Simulator, settled seconds: x50 step 62 to 66 (Kaspa 1,542), /50 step 657 to 753 (Kaspa 12,296). Apple M5 Max under load 7 to 442; the DAG model is fitted on one scale; the pool hopper's 0.7-point excess over Kaspa's rule stays opennone yet 13Every node executes the ordered transactions natively and reaches the same state root
Litepaper Proving ("Every node executes ... natively"), Building ("runs on Igneum unchanged")
tested by the teamrepo f5f8c80, 8dae48b; fork devnet-v4 dc749905; revm 43.0.3node tools/evm-smoke/smoke.mjs against a 3-node igneumd; igneum-exec-diff seq.json; bench-log "execution layer devnet v3" and "devnet-v4 integration"Simnet, 3 October 2026: 87 of 87 viem checks, state roots identical on 3 nodes at four heights, 57 executed and 19 skipped transactions agree with plain revm, 0 mismatches. Merged node on real proof of work, 4 October 2026: 84 of 85 checks (the miss needs parallel blocks the network did not produce in 36 s), 59 transfers in 10 chain blocks, state roots identical on 3 nodes, igneum-exec-diff 0 mismatches over 59 transactions; the live devnet v4 runs this execution layer. Apple M5 Max. The prover is a stub; state is rebuilt from genesis at start; no EVM transaction relay between nodesnone yet 14Ethereum bytecode runs unchanged, with the documented differences of spec 7.1
Homepage Build card; litepaper Building
tested by the teamas row 13; fixes F-exec-A, F-exec-B (spec 7.5)tools/evm-smoke/smoke.mjs: deploy via viem, increment, hashLoop, eth_estimateGas, eth_getLogs; tools/exec-attacks scenarios 1 and 3; bench-log "execution layer attack fixes"Deployment, calls, reverts, logs and gas estimates behave as viem expects; chain id 4463 at that version (4464 since the class v5 floor, 8 October 2026); the prototype pgas table gives 0.0095 to 0.028 pgas per gas, below the design's band before calibration, 3 October 2026. 4 October 2026: a transaction that would cross the block's proving budget is refused by the mempool and, if forced in, aborted and charged with its nonce advanced (25 of 25 checks; 30 of 30 malformed cases). Apple M5 Max. The Prover precompile, proof records and the shard planner are not in the nodenone yet 15Every block is proven, with the proof landing within about a minute at launch
Homepage stats ("~60 s to a proof"); litepaper Proving; roadmap phase 3 gate
implementedrepo d7e1f89 (GPU proof), e01a3cc, 292e800, eedd136 (proving/igneum-prove: shard cutter, MPT witnesses, shard and aggregator guests); SP1 6.8.1; spec 7.2, 7.6proving/windows-wsl2 (SETUP-PROVER, PROVE-BLOCK) on the RTX 5090; igneum-prove-host --mode block on proving/fixtures/; bench-log "proving v0 on the RTX 5090" and "proving: devnet v4 shards"First GPU proof of an Igneum block, 4 October 2026, RTX 5090 (WSL2, SP1 cuda, mining paused): fixture block-78-increment (2 transactions), core proof 1.4 s (7.3 MB, verify 0.221 s), compressed proof 2.7 s (1.27 MB, verify 0.038 s), post-state and receipts roots identical to the node's; 15.7x and 20.6x faster than a loaded M5 Max CPU. The same day on that CPU (load 38 to 47): a three-shard block proved shard by shard and aggregated by recursion, 19 min (1,139 s) end to end, 245 to 337 s per compressed shard proof, every proof verified. What is not there: no proof is produced, carried or checked on the chain (the devnet prover is a stub that signs claims), the proving pool pays nobody (row 21), the block proven is far below one shard, and the 60-second figure remains a design target; the pass mark is the standard in docs/benchmarks/proving-e2e.md. Second RTX 5090 run, 4 October 2026 evening (job run-20261004-173115): a full shard at the provisional S_p (6.75 M pgas, 60.8 M cycles) executed in 1.63 s, core proof 8.3 s (18.1 MB), compressed proof 10.9 s (1.27 MB, verify 0.040 s); a two-shard block (13.5 M pgas) proved shard by shard (11.7 s and 10.0 s) and aggregated in 2.2 s, 24 s of GPU stages end to end, every proof verified, six tampered witnesses rejected. The two host defects (an abort after the upload, an idle wait that turned out to be an unbuffered 18 MB proof save through the WSL2 file bridge, 24 minutes) are fixed (ledger P20) 5 October 2026, live devnet with real transactions (bench-log "real transactions, the first non-empty shard proven and paid"): block 72704 shard 0, 29 transfers, 5,800 pgas, proven on the RTX 5090 Windows rig in 34 s, verified on the Apple M5 Max in 0.297 s and paid 1.7623 IGN, 53 s after the chain block executed; of about 1,400 blocks in the 20-minute window 36 were proven (the one prover takes the newest shard assigned to it), so "every block" is not yet true; a second content shard (72803, all copies skipped) failed the native-execution veto on the exporter's block structure, fixed with fixtures the same day, the node side pending the 0.3.9 rollout 5 October 2026, evening (bench-log "proving v1"): the aggregated segment record, the chain rule and the unproven rule are implemented behind proving_v1_activation_daa (branch proving-v1, not on the devnet before 0.3.11); on the RTX 5090 a chain of 8 consecutive live blocks proved and aggregated by recursion in 135.6 s with the miner on the card (17 s a block, one proof of 1,272,909 bytes attesting all 8, verified in 0.04 s); the 3-node fast-time harness paid a segment record 1.0 s after submission and refused a late one after its deadline (21 checks); the devnet itself, with one prover, carried proofs for 2.4% of blocks over 30 minutes at a block-to-record latency p50 44 s, p99 52 s. The "within about a minute" holds per proven block; "every block" needs 18 mining 5090s or 6 proving-only cards at empty blocks on the measured rates, and the mandatory rule stays off until the share is onenone yet 16A 12 GB card proves one shard in about 20 s (WITHDRAWN 5 October 2026: a 24 GB card proves a full shard at the adopted size in 4.3 s; 32 GB mines and proves)
Litepaper Proving ("The proving budget"); roadmap gate 2
designedspec 5.1 (Target), 7.6 (S_p provisional, 7,500,000 pgas = B_p / 4)PROVE-SHARD.bat on the RTX 5090 (pending); the end-to-end standard in docs/benchmarks/proving-e2e.md; bench-log "proving: devnet v4 shards"Measured on a 32 GB card, not yet on a 12 GB card. A shard at the provisional S_p is 60.8 M SP1 cycles on the prototype pgas table (9 cycles per pgas, 44 per EVM gas; the modexp entry about 100x its SP1 cost); on an RTX 5090 (4 October 2026 evening, job run-20261004-173115) it executed in 1.63 s and its compressed proof took 10.9 s, verified in 0.040 s, so the 32 GB card is inside the 20 s target with margin. Whether a 12 GB card proves it at all, and in what time, is the next measurement (an RTX 3060 and an RTX 5060 Ti 16 GB are on order). A per-shard time can be met by shrinking the shard, so the project does not use it as a pass mark 5 October 2026, evening (bench-log "proving v1", the S_p curve): measured on the RTX 5090 with SP1 6.8.1's GPU prover, the card to itself, 1-s nvidia-smi samples: an empty shard 13,874 MiB and 2.2 s; a full shard at the ADOPTED v1 budget (30,000 pgas, 4.7 M cycles) 20,434 MiB and 4.3 s; the full prototype shard (6.75 M pgas, 60 M cycles) 28,307 MiB and 10.8 s; beside the miner 15,670 and 30,039 MiB. No environment knob of SP1 moves the 13.9 GB floor and the GPU server has no options of its own, so on this build a 12 GB card proves nothing, a 16 GB card only empty shards, a 24 GB card the adopted full shard alone and beside the miner (22,210 MiB and 13.2 s, measured on the 32 GB card: the 5090's allocation pattern, not yet a run on a 24 GB card) and a 32 GB card the prototype shard beside the miner with 2.5 GB spare. The litepaper line now says so; the 12 GB gate returns when a prover build with a smaller floor is measured on a 12 GB cardnone yet -17The chip resistance claim, served as the class v6 close words it, the claim statement above the sentence: Igneum remains competitive on accessible commodity GPUs even when specialised mining hardware is assumed to exist, remain compatible and seek profit; its security does not rely on identifying that hardware or retiring it through emergency changes. Class v6 adopts the 64-register window and retains it across every rotation. Current modelling places the strongest specialised designs assessed against the GPU tier at about 2.3x to 3.3x energy-efficiency advantage a node ahead (2.0x to 2.9x on the GPU's own node), a bracket that is approximate and provisional until the placed gated core rows land. The long-program and select-tree proposals were rejected. Economic resistance depends on development cost, deployment economics and productive hardware lifetime; family transitions receive an obsolescence benefit only where a loss of competitiveness is demonstrated; programmable multi-epoch designs are included in the assessment. Beside it: class v4 is live from the first block on the testnet and the mainnet; class v5 makes the dataset the chain's state so a stateless or stale chip is wrong on every item; the hot-set cache is bounded at 1.067x at the ceiling and the weak-day FPGA at 12 percent on 15 days a century, both routed to the next class; datacentre silicon does not change the question; the three statements (energy resistance, economic resistance, response capability) are served separate, with the harness and the scoring rule linked
the home page's chip line, the litepaper's chip section (/litepaper#chip-model), the miner page's line
the GPU side tested by the team (the RTX 5080 and RTX 5090 clock-lock passes under class v4, every card, the verifier, the two attack-pass bounds, the H100); the chip core synthesised on ASAP7 and scaled to N3, claimed; the chip's memory modelled; the node column claimed scaling; the economic surface modelled, first cut, conditional; the rotation schedule measured per boundary; class v5 designed; the Antminer X5 an observed comparison, not a ceiling; the X9 a withdrawn pre-order, never benchmarkeddocs/design/class-v6-rotating-family.md section 10 (10.0 to 10.0h, the close and its two accepted external reviews, 8 October 2026); docs/design/class-v5-stored-state.md sections 0, 13 and 14 and docs/design/class-v5-harness/ (branch class-v5); docs/design/class-rotation-four-layers.md; docs/analysis/chip-model-v3.md 5 and 6; docs/analysis/latency-shadow-2026-10-06.md; docs/analysis/attack-pass/f8-uniform.md, f4-weakday.md, docs/analysis/ca3-v4-uniform.md; the H100 row of 7 October; docs/plans/cryptanalysis/in-house-pass.md (the internal adversarial pass)the scoring rule in the close (the minimum over workloads of the maximum over free adversarial designs of the GPU's joules per hash over the adversary's, under the 10 percent GPU-cost budget at the lock, the verifier limit, cross-vendor correctness and hardware accessibility); the card rows by the benchmark package; the class v5 harness and the family harness; the attack-pass harnesses tools/attack/f8-uniform and the F4 census; the verifier by igneum-pow bench136 MH/s at 350 W (5090, bench) and 290 W (app); the class v4 efficiency passes (bench log "7 to 8 October 2026, the class v4 efficiency passes: the core clock lock on the RTX 5090 and the RTX 5080", measured): the 5090 at 136.84 MH/s and 475.5 W unlocked, 134.98 at 316.3 W at a 1,400 MHz core lock, the best points class v4 at 1,200 MHz (133.80 MH/s, 305.1 W, 0.439 MH/W) and class v3 at 1,300 MHz (134.62, 223.3 W, 0.603), the premium 145 W unlocked and 82 W at the best points, the knee 1,300 MHz; the RTX 5080 (8 October 2026, the dock card of the three-card Windows rig) at 71.41 MH/s and 253.1 W unlocked under class v4 against 71.28 at 169.7 W under class v3, the best points class v4 at 1,100 MHz (71.20 MH/s, 146.6 W, 0.486 MH/W) and class v3 at 1,000 MHz (71.11, 103.7 W, 0.686), the premium 83.4 W unlocked and 41 W at the best points, the knee between 1,000 and 900 MHz; per tier: a 5080 owner on class v4 locked near 1,100 MHz pays 147 W instead of 253 for 0.3 percent less rate, MH per watt up 72 percent, the lever Ember Tune's core-clock knob in 0.3.24; 27 MH/s at 21 W (M5 Max); 249 MH/s (H100 SXM) at 98 percent of its read ceiling, 1.78x hash, 1.15x MH/W, a third per rented dollar; 2.33 ms per warp; the GPU side of the close: the RTX 5080 at its 1,100 MHz lock 2.06 microjoules per hash and the RTX 5090 at its 1,300 MHz lock 2.33 (class v4, 8 October 2026); the modelled bracket about 2.3x to 3.3x a node ahead and 2.0x to 2.9x node for node (approximate, provisional; the clock-gated base core k about 0.37 at N3, the gated window adding about 0.13; the first placed core 64 percent above synthesis), the placed gated row expected near 2.6x to 3.1x and 2.3x to 2.7x and served when it lands; the shadow's premium on a 5090 81.8 W at its knee (class v4 at the 1,200 MHz lock 133.80 MH/s at 305.1 W; class v3 at 1,300 MHz 134.62 at 223.3 W; 7 October 2026); 1.067x at the ceiling; 12 percent on 15 days a century; 10.85 ms at rung 3; 6 to 8 October 2026, the M5 Max, the desk rigs' RTX 5090, RTX 5080, RX 9070 XT and RTX 4070, rented pods, a rented H100 SXM, igneum-build-1 Bitmain's Antminer X9 (RandomX; 1,000 KH/s, 2,472 W, 2.47 J per KH, USD 5,600; pre-orders 26 December 2025) was withdrawn in mid-May 2026 with buyers refunded before any unit shipped, no independent benchmark, commodity Sophgo SG2044 server SoCs with an AES accelerator, no tapeout: it is a precedent on the served pages, not a chip core against the model (attack pass AP-F5-1, 7 October 2026; the close's 10.0f, 8 October 2026). Withdrawn from the served pages on 8 October 2026 and not restated: the 2.1x and 3.4x launch line, the 5x to 9x class v3 baseline, the ladder's 2.8x row, the USD 100 M pay-back row, any chip-arrival probability, the lifetime claim and the USD 300 M and 340 M lines.none yet; the next test is the internal adversarial pass (three lanes new to the hash code, outsider inputs only, reports published whole), whose floor reading is in (measured, 8 October 2026; ledger AP-F8-1 and AP-F8-6): nine of nine hot sets refused; the diffuse era-stride excess, bounded under 0.1 percent of a hash's reads per site, is not caught by the floor and is the next class's test, and no outside review has run yet +17The chip resistance claim, served as the class v6 close words it, the claim statement above the sentence: Igneum remains competitive on accessible commodity GPUs even when specialised mining hardware is assumed to exist, remain compatible and seek profit; its security does not rely on identifying that hardware or retiring it through emergency changes. Class v6 adopts the 64-register window and retains it across every rotation. Current modelling places the strongest specialised designs assessed against the GPU tier at about 2.3x to 3.3x energy-efficiency advantage a node ahead (2.0x to 2.9x on the GPU's own node), a bracket that is approximate and provisional until the placed gated core rows land. The long-program and select-tree proposals were rejected. Economic resistance depends on development cost, deployment economics and productive hardware lifetime; family transitions receive an obsolescence benefit only where a loss of competitiveness is demonstrated; programmable multi-epoch designs are included in the assessment. Beside it: class v5 derives the dataset from chain state; whether that excludes a specialised design is under evaluation (Deliverable 3), since a design that tracks state is not excluded by staleness; the hot-set cache is bounded at 1.067x at the ceiling and the weak-day FPGA at 12 percent on 15 days a century, both routed to the next class; datacentre silicon does not change the question; the three statements (energy resistance, economic resistance, response capability) are served separate, with the harness and the scoring rule linked; the energy ratio is not the pass criterion: the coexistence model ([docs/analysis/class-v6/coexistence-model.md](https://git.igneum.network/igneum-network/igneum/src/branch/master/docs/analysis/class-v6/coexistence-model.md)) is, and its first run's result is served with its conditions; the window sentence and the bracket stay provisional until the placed gated rows land
the home page's chip line, the litepaper's chip section (/litepaper#chip-model), the miner page's line
the GPU side tested by the team (the RTX 5080 and RTX 5090 clock-lock passes under class v4, every card, the verifier, the two attack-pass bounds, the H100); the chip core synthesised on ASAP7 and scaled to N3, claimed; the chip's memory modelled; the node column claimed scaling; the economic surface modelled, first cut, conditional; the rotation schedule measured per boundary; class v5 designed; the Antminer X5 an observed comparison, not a ceiling; the X9 a withdrawn pre-order, never benchmarkeddocs/design/class-v6-rotating-family.md section 10 (10.0 to 10.0h, the close and its two accepted external reviews, 8 October 2026); docs/design/class-v5-stored-state.md sections 0, 13 and 14 and docs/design/class-v5-harness/ (branch class-v5); docs/design/class-rotation-four-layers.md; docs/analysis/chip-model-v3.md 5 and 6; docs/analysis/latency-shadow-2026-10-06.md; docs/analysis/attack-pass/f8-uniform.md, f4-weakday.md, docs/analysis/ca3-v4-uniform.md; the H100 row of 7 October; docs/plans/cryptanalysis/in-house-pass.md (the internal adversarial pass)the scoring rule in the close (the minimum over workloads of the maximum over free adversarial designs of the GPU's joules per hash over the adversary's, under the 10 percent GPU-cost budget at the lock, the verifier limit, cross-vendor correctness and hardware accessibility); the card rows by the benchmark package; the class v5 harness and the family harness; the attack-pass harnesses tools/attack/f8-uniform and the F4 census; the verifier by igneum-pow bench136 MH/s at 350 W (5090, bench) and 290 W (app); the class v4 efficiency passes (bench log "7 to 8 October 2026, the class v4 efficiency passes: the core clock lock on the RTX 5090 and the RTX 5080", measured): the 5090 at 136.84 MH/s and 475.5 W unlocked, 134.98 at 316.3 W at a 1,400 MHz core lock, the best points class v4 at 1,200 MHz (133.80 MH/s, 305.1 W, 0.439 MH/W) and class v3 at 1,300 MHz (134.62, 223.3 W, 0.603), the premium 145 W unlocked and 82 W at the best points, the knee 1,300 MHz; the RTX 5080 (8 October 2026, the dock card of the three-card Windows rig) at 71.41 MH/s and 253.1 W unlocked under class v4 against 71.28 at 169.7 W under class v3, the best points class v4 at 1,100 MHz (71.20 MH/s, 146.6 W, 0.486 MH/W) and class v3 at 1,000 MHz (71.11, 103.7 W, 0.686), the premium 83.4 W unlocked and 41 W at the best points, the knee between 1,000 and 900 MHz; per tier: a 5080 owner on class v4 locked near 1,100 MHz pays 147 W instead of 253 for 0.3 percent less rate, MH per watt up 72 percent, the lever Ember Tune's core-clock knob in 0.3.24; 27 MH/s at 21 W (M5 Max); 249 MH/s (H100 SXM) at 98 percent of its read ceiling, 1.78x hash, 1.15x MH/W, a third per rented dollar; 2.33 ms per warp; the GPU side of the close: the RTX 5080 at its 1,100 MHz lock 2.06 microjoules per hash and the RTX 5090 at its 1,300 MHz lock 2.33 (class v4, 8 October 2026); the modelled bracket about 2.3x to 3.3x a node ahead and 2.0x to 2.9x node for node (approximate, provisional; the clock-gated base core k about 0.37 at N3, the gated window adding about 0.13; the first placed core 64 percent above synthesis), the placed gated row expected near 2.6x to 3.1x and 2.3x to 2.7x and served when it lands; the shadow's premium on a 5090 81.8 W at its knee (class v4 at the 1,200 MHz lock 133.80 MH/s at 305.1 W; class v3 at 1,300 MHz 134.62 at 223.3 W; 7 October 2026); 1.067x at the ceiling; 12 percent on 15 days a century; 10.85 ms at rung 3; 6 to 8 October 2026, the M5 Max, the desk rigs' RTX 5090, RTX 5080, RX 9070 XT and RTX 4070, rented pods, a rented H100 SXM, igneum-build-1 Bitmain's Antminer X9 (RandomX; 1,000 KH/s, 2,472 W, 2.47 J per KH, USD 5,600; pre-orders 26 December 2025) was withdrawn in mid-May 2026 with buyers refunded before any unit shipped, no independent benchmark, commodity Sophgo SG2044 server SoCs with an AES accelerator, no tapeout: it is a precedent on the served pages, not a chip core against the model (attack pass AP-F5-1, 7 October 2026; the close's 10.0f, 8 October 2026). Withdrawn from the served pages on 8 October 2026 and not restated: the 2.1x and 3.4x launch line, the 5x to 9x class v3 baseline, the ladder's 2.8x row, the USD 100 M pay-back row, any chip-arrival probability, the lifetime claim and the USD 300 M and 340 M lines.none yet; the next test is the internal adversarial pass (three lanes new to the hash code, outsider inputs only, reports published whole), whose floor reading is in (measured, 8 October 2026; ledger AP-F8-1 and AP-F8-6): nine of nine hot sets refused; the diffuse era-stride excess, bounded under 0.1 percent of a hash's reads per site, is not caught by the floor and is the next class's test, and no outside review has run yet 18The chip resistance measurements: the program is latency-bound (dependent reads spread over the whole dataset), not bandwidth-bound, on every card we own, and sits beyond a card's on-chip cache; measured 8 October 2026: the distinct-index floor holds at 0.995 on every accepted program, about half of epochs carry one load site with a biased address bit at the era's stride rotation, priced at about 1.6 percent of a hash's reads to a chip storing half the dataset and nothing to one storing all of it (docs/analysis/class-v6/family-gate.md, lane D; adv-cache-2's report-chained-cache-2.md section 2.3 on its branch; ledger AP-F8-7)
Litepaper Mining ("waits on memory latency, not on maths or bandwidth"), vs RandomX; the numbers page
tested by the teamreadwidth e752fc7 (docs/plans/read-width.md), ca2-era 78c0ee4, ca2-cache 2de19e5 (docs/plans/hot-table.md)The dependent-read probes at 32 to 1,024 MiB and the hash rate per class on the three cards; the latency-bound share = rate over the probe ceiling per loadLatency-bound share at the 1 GiB dataset: RTX 5090 0.96 (v2) and 1.01 (v3), RX 9070 XT 0.87 and 0.95, M5 Max 1.01 and 1.06; wider reads do not close the AMD gap (the 9070 XT does 2.4 G dependent reads per second at every width; the 5090 goes bandwidth-bound at 64 B, share 0.58); a 32 to 96 MiB hot table is not kept resident by any card while the dataset streams (g 0.80 to 0.87 in the added form). 5 October 2026none yet 19The lottery hash is sound as a hash: uniform output, deterministic, no out-of-bounds read, fuzzed; class v3 bit-exact on the three vendors
Litepaper vs RandomX ("Every number above is measured and logged"), the numbers page
tested by the teamca2-mixer 1ab8b21 (tests/mixer.rs, tests/scratch.rs), ca2-era 78c0ee4, ca2-soundness a465881 (docs/analysis/scratch-soundness.md), igneum-pow/tests/packs.rsThe crate suite (53 + 4 + 19 + 7), the Metal fuzz, edge, stats and determinism runs on the v3 construction, the pack vectors and 2^24 fingerprints on Metal, Apple OpenCL, the RTX 5090 and the RX 9070 XT, the 1,024-hash CPU re-check per cardClass v3 (mixer x8 + era): 200-program fuzz 200 of 200 on Metal, every tenth on Apple OpenCL; the pinned v3 packs 3/3 + 3/3 and 96 of 96 lanes on Metal and Apple OpenCL; the six era packs' fingerprints equal on the three vendors (the three-card Windows rig (RTX 5090, RTX 4070, RX 9070 XT) job run-ca2-era-pc1-20261005, 5 October 2026); the v2 exports byte-identical on the v3 crate; the final-class PC rows and the G2 re-check: job run-ca2-era-pc1b-20261005 (pending at the time of writing)none yet 20No premine, no pre-sale, no allocation: every coin is minted by the schedule and every coin goes to the block producer (80%) and the proving pool (20%)
Homepage stats and Economics tiles; litepaper Supply, Economics
implementedrepo 6ac80a3; fork "igneum-node devnet v0"; consensus/core/src/igneum.rs, coinbase.rscargo test -p kaspa-consensus-core igneum (8 pass: subsidy table, ramp, split, cap) and cargo test -p kaspa-consensus coinbase (8 pass); igneum-miner inspect 40; bench-log "igneum-node devnet v0"Coinbases on the devnet: 80/20 exact on 39 of 39 single-payee blocks, the 20% to the igneum-proving-pool-v0 output; the per-second schedule sums to under the 4,000,000,000 cap by less than 100 coins; 3,168,808,781 units per DAA second in years 0 to 2, halving at 63,115,200 DAA s. 3 October 2026, Apple M5 Max. The devnet genesis carries no allocation; the mainnet genesis does not exist yet, so the claim is about the code and the stated rule, not a launch that has happenednone yet @@ -296,7 +296,7 @@ td.mono{font-family:var(--f-mono);font-size:12.5px;min-width:180px}td.iv{color:v 30One click: install, press start, the card mines; the app looks after its node
Homepage Mine section ("One click: install, press start"); litepaper "One click, for everyone else"; journey phase 5
tested by the teamrepo 3bb50d6, 2c4b30f, 6461540 (package 0.3.0: prebuilt NVRTC CUDA worker and generic OpenCL worker, driver only), a1a33cb, 7c794df, 0d4498e, 6c083db (Igneum Miner 0.3.0), 78903cd (0.3.1, over-the-air updates)Igneum-Miner-Setup-0.3.0.exe (runner-built, unsigned) on a an RTX 5090 on Windows with an RTX 5090 and no toolchain; proto-cuda/nvrtc/emu/serve-check.sh on the Apple M5 Max; proto-cuda/windows-app/TEST.md; bench-log "one-click Windows workers", "first machine on the Igneum Miner app", "a node 60 s behind the clock is silently dead", "the gfx1036 worker fault"Four machines by 14:45 UTC on 4 October 2026: the RTX 5090 Windows rig, then the three-card Windows rig (RTX 5090, RTX 4070, RX 9070 XT) (RTX 5090 at 110 MH/s under the 80% power cap), the project's Apple M5 Max (25 MH/s) and the outside Apple silicon laptop (row 29), all on Igneum Miner 0.3.1. The NVRTC worker compiled the pack on the card with no toolchain installed and mined at 124.2 MH/s, equal to the nvcc-built worker, 0 rejected, CPU re-check clean; inside the app 117 to 119 MH/s with 34 accepted blocks in the first minute, the integrated AMD chip at 3.3 MH/s beside it (row 9). Two defects found by the install, both fixed the same hour: a clock 62 s slow after a power cut made the node reject every relayed block for 12 minutes with no visible reason (the app now reads the skew from the node's warnings, the block timestamps over the EVM RPC and an HTTPS Date header, warns over 5 s and blocks Start over 10 s, with a one-click clock sync; checked on the Apple M5 Max with a fake 60 s skew; a one-line node warning is filed), and the node card said "syncing" while the miner was already accepted. The Mac could only emulate the NVIDIA path (17 of 17 sampled hashes) and the AMD path on Apple OpenCL (15 of 15). Over-the-air updates were dry-run on a private devnet (0.3.0 to 0.3.1 and back), not on a user's machine. The installer is unsigned (SmartScreen "run anyway"). Second machine, the same afternoon: a friend of the project installed Igneum Miner 0.3.1 from the DMG on an Apple silicon laptop with no toolchain and no instructions beyond five steps; the node synced from the seed, the Metal worker reported ready, 33 accepted blocks and 0 rejected in 7 minutes at 21.0 MH/s average, CPU re-check OK on every share, uploads arriving every minute under its per-install id. That laptop is not the project's hardware, but the result is observed through the project's own log intake and reported by the project, so it stays tested by the team until an outsider publishes a run of their own. The devnet's other GPU machines (the three-card Windows rig (RTX 5090, RTX 4070, RX 9070 XT) and the Apple M5 Max) run the same workers through the launcher, not the appnone yet 31Card lifetime: a 4 GB card mines about four years and an 8 GB card about twelve, under the dataset's step schedule (2 GB at genesis, doubling at years 4, 12, 28, 60) with the cache freed after the daily build
Litepaper Hardware and vs RandomX ("Dataset" row); homepage Mine card and "Memory" row
designeddocs/analysis/card-lifetime-2026-10-05.md (branch card-lifetime 1fecfe2); spec 1.13.3 option (b) recommended to the owner 5 October 2026 (docs/plans/counter-asic-2-rollout.md 6c)The per-tier working-set arithmetic of that document (GTX 1650, RTX 3050, RTX 3060, RTX 4090 tiers) against the step scheduleA design claim: under the continuous mapping (a) a 4 GB card is out within 1 to 1.5 years and an 8 GB card at 6 to 7.5 years, so the sentence is true only under the step schedule (b), which the spec has not yet fixed (O-1.13)none yet -

Click a column heading to sort; click again to reverse. Versions: igneum-pow is the Rust crate at version 0.2.0 (generator version 2, 4 October 2026); Devnet 3 runs node f8da7515 on release-0.3.25-node with igneum-pow 1c420786, chain id 4464 since its class v5 floor (4463 from genesis), read 8 October 2026, 15:50 UK; the current state, machine-readable, is the release manifest, which fills these at build time. The labels stay distinct: a claim never moves up a label without the artefact the next table names. The public reference repository exists now (the specifications, the pow crate, the simulators and the test material, at git.igneum.network); the full node, the miner and the proving code open later, so a row that cites them is tested by the team until they do. Repo commits are this repository's; fork commits are the node fork and its worktrees, named by message as the engineering log names them. Source of every number: the engineering log. The source of this page is docs/evidence.md in the repository.

+

Click a column heading to sort; click again to reverse. Versions: igneum-pow is the Rust crate at version 0.2.0 (generator version 2, 4 October 2026); the devnet runs node f8da7515 on release-0.3.25-node with igneum-pow 1c420786, chain id 4464 since its class v5 floor (4463 from genesis), read 8 October 2026, 15:50 UK; the current state, machine-readable, is the release manifest, which fills these at build time. The labels stay distinct: a claim never moves up a label without the artefact the next table names. The public reference repository exists now (the specifications, the pow crate, the simulators and the test material, at git.igneum.network); the full node, the miner and the proving code open later, so a row that cites them is tested by the team until they do. Repo commits are this repository's; fork commits are the node fork and its worktrees, named by message as the engineering log names them. Source of every number: the measurement record in the repository (docs/bench-log.md). The source of this page is docs/evidence.md in the repository.

What would move a row

diff --git a/site/litepaper.html b/site/litepaper.html index 12639b33a..d7e0411f0 100644 --- a/site/litepaper.html +++ b/site/litepaper.html @@ -4,7 +4,7 @@ Igneum Litepaper: how the chain works - + @@ -295,14 +295,14 @@ body.all .pager{display:none}
Igneum/The litepaper
-
The litepaper · version 0.2
+
The litepaper · Igneum 2.0

Mined by GPUs.
Proven by fire.

-
A proof-of-work chain whose miners also prove every block, run Ethereum's apps, and are protected from specialised chips by a program that changes every hour.
+
A GPU-secured network for Ethereum-compatible applications and verifiable computation.
- Published 3 October 2026 · updated 7 October 2026 + Published 3 October 2026 · updated 8 October 2026 Coin IGN · cap 4,000,000,000 - Status Devnet 3 live, testnet armed + Status The Igneum 2.0 devnet is starting Method one founder with AI systems · external review before gate 3 This is not an offer to sell anything
@@ -339,11 +339,11 @@ body.all .pager{display:none}

Abstract

-

Igneum is a proof-of-work blockchain built for graphics cards, where NVIDIA cards also prove every block with zero-knowledge proofs and sell proving to other chains. Igneum remains competitive on accessible commodity GPUs even when specialised mining hardware is assumed to exist, remain compatible and seek profit; its security does not rely on identifying that hardware or retiring it through emergency changes. Class v6 adopts the 64-register window and retains it across every rotation. Current modelling places the strongest specialised designs assessed against the GPU tier at about 2.3x to 3.3x energy-efficiency advantage a node ahead (2.0x to 2.9x on the GPU's own node), a bracket that is approximate and provisional until the placed gated core rows land. The long-program and select-tree proposals were rejected. Economic resistance depends on development cost, deployment economics and productive hardware lifetime; family transitions receive an obsolescence benefit only where a loss of competitiveness is demonstrated; programmable multi-epoch designs are included in the assessment. The chip model: every number labelled measured, modelled or claimed, the harness and the scoring rules beside it.

-

It runs the Ethereum virtual machine, so anything built for Ethereum runs on Igneum unchanged. Transactions are included in about one second, proven within about a minute at launch, and locked by miners within about two. There is no premine, no pre-sale, no treasury taken from emission, no stake anywhere in consensus, and no dependence on any other chain. Mining stays open to anyone with a GPU because the mining program changes every hour, so a chip built for one program is useless for the next, and a chip for the whole program space is a GPU without the graphics parts. No scheduled human release is needed to keep it that way. Writing new code, including an emergency fix to the proof system, is the one thing that takes a person, and it activates only on miner signalling.

+

A GPU-secured network for Ethereum-compatible applications and verifiable computation. Igneum is a proof-of-work chain built for graphics cards. AMD, Apple and NVIDIA cards mine; NVIDIA cards also prove its blocks with zero-knowledge proofs. Igneum remains competitive on accessible commodity GPUs even when specialised mining hardware is assumed to exist, remain compatible and seek profit; its security does not rely on identifying that hardware or retiring it through emergency changes. Class v6 adopts the 64-register window and retains it across every rotation. Current modelling places the strongest specialised designs assessed against the GPU tier at about 2.3x to 3.3x energy-efficiency advantage a node ahead (2.0x to 2.9x on the GPU's own node), a bracket that is approximate and provisional until the placed gated core rows land. The long-program and select-tree proposals were rejected. Economic resistance depends on development cost, deployment economics and productive hardware lifetime; family transitions receive an obsolescence benefit only where a loss of competitiveness is demonstrated; programmable multi-epoch designs are included in the assessment. The chip model: every number labelled measured, modelled or claimed, the harness and the scoring rules beside it.

+

It runs the Ethereum virtual machine through revm, so contracts built for Ethereum deploy with familiar tools; the differences (block context, randomness, two-dimensional fees) are documented. Transactions are included in about one second, proven within about a minute at launch, and locked by miners within about two (designed targets). There is no premine, no pre-sale, no treasury taken from emission, no stake anywhere in consensus, and no dependence on any other chain. Mining is built to stay open to anyone with a GPU. The argument rests on the scoring rule's result against placed adversary designs and on the economics, reported separately as energy advantage, economic advantage and response capability. The hourly program is an optional improvement, not the defence. Writing new code, including an emergency fix to the proof system, is the one thing that takes a person, and it activates only on miner signalling.

-
1 / s
blocks, rising to 10
-
~60 s
to a proof at launch
+
1 / s
blocks, rising to 10 (designed)
+
~60 s
to a proof at launch (designed)
0
premine or stake
100%
to miners and provers
@@ -353,12 +353,12 @@ body.all .pager{display:none}

Precedents, and what Igneum adds

Every piece of Igneum has a precedent somewhere. We know of no chain that combines them. The table names the closest precedent for each piece, what Igneum adds, and how far each piece has got. It will be corrected when shown wrong.

FromToWhat it takes
designedimplementedCode in this repository with test vectors that pass
- + - - - - + + + + @@ -392,7 +392,7 @@ body.all .pager{display:none} 1. Mining lotteryA random GPU program picks who makes the next block - New program every hour: a chip wired for one program is useless + A new program every hour; the chip model scores the rest @@ -404,12 +404,12 @@ body.all .pager{display:none} 3. EVM execution Blocks carry transactions only; the proof computes the state - Solidity, wallets and tooling work unchanged + Familiar Solidity, wallets and tooling; differences documented 4. Miners prove the block - Shards proven on consumer GPUs, aggregated into one proof + Shards proven on NVIDIA cards, aggregated into one proof Lands within about a minute at launch, paid from gas @@ -421,34 +421,34 @@ body.all .pager{display:none} External proving jobs - Rollups and bridges pay Igneum - Same GPUs, same proof format + Designed: rollups and bridges buy proofs + Same NVIDIA cards, same proof format winning blockordered blocksstate transitionsaggregated proof
Five layers plus the external proving market. A block flows down the column; outside demand feeds the same miners from the side.
-

Live on Devnet 3, 7 October 2026

-

Devnet 3 (igneum-devnet-4, chain id 4464 since its class v5 floor at DAA 68,400, 4463 from genesis to the floor; the chain’s current state is the release manifest) made its first block at 18:06 UK on 7 October 2026 with every upgrade on from block zero, and locked its first checkpoint at 20:02 UK. The first devnet ran from 3 October 2026 and took each upgrade by height. Coins on Devnet 3 have no value and the chain may be reset. What is on it:

+

Live on the devnet

+

The Igneum 2.0 devnet is the network. Its coins have no value and the chain may be reset. What is on it:

PieceClosest precedentWhat Igneum addsState, 7 Oct 2026
PieceClosest precedentWhat Igneum addsState, 8 Oct 2026
A mining program that regenerates itself, for GPUsRandomX on Monero since 2019, for CPUs, a program per hash; one chip shipped against it, Bitmain’s Antminer X5 (September 2023), a board of RISC-V chips at 1.46x per joule over a desktop CPU, on silicon believed mining privately from about 2021; the X9 was withdrawn in May 2026 with zero units; RandomX v2 was released on 25 March 2026 with its activation pending. ProgPoW, as KAWPOW on Ravencoin since 2020, changes the maths inside a fixed program shape every few blocks on GPUs (approximate)A whole kernel per hour compiled to native code, a daily dataset from a 256 MB cache, a verifiable delay before the seed, era draws from a genesis reserveMeasured: hourly swaps on Apple, NVIDIA and AMD cards on the live devnet, 4 October 2026
The mining card does paid, useful, verifiable workPrimecoin's prime chains in 2013 were not useful. Aleo ran proving as consensus and the fastest prover won (both approximate)Proving kept apart from the lottery; shards assigned by sortition, not by speedImplemented: proving v0 and v1 on Devnet 3 from block zero (7 October 2026), v0 on the first devnet since 5 October 2026. The job market for other chains is Designed
A proof-of-work chain where every block is provenzkEVMs run as rollups on proof-of-stake Ethereum. Conflux has run GPU-mined EVM apps on a DAG since 2020, without proofs (approximate)Proven state on a proof-of-work base layer, produced by the miners themselvesImplemented in part: shards proven and paid on the devnet. The aggregated block proof checked in consensus is Designed
Finality held by miners and not moved by hour-long rentalsDecred votes with stake. Horizen penalises hidden chains. Kaspa limits merge depth (approximate)Vote weight is 30 days of blocks per key. Hashrate that appeared today has no voteImplemented: rule v3 live on Devnet 3 from block zero, first lock 7 October 2026; rule v2 ran the first devnet from its first lock on 4 October 2026. External review is owed at gate 3
A mining program that regenerates itself, for GPUsRandomX on Monero since 2019, for CPUs, a program per hash; one chip shipped against it, Bitmain’s Antminer X5 (September 2023), a board of RISC-V chips at 1.46x per joule over a desktop CPU, on silicon believed mining privately from about 2021; the X9 was withdrawn in May 2026 with zero units; RandomX v2 was released on 25 March 2026 with its activation pending. ProgPoW, as KAWPOW on Ravencoin since 2020, changes the maths inside a fixed program shape every few blocks on GPUs (approximate)A whole kernel per hour compiled to native code, a daily dataset from a 256 MB cache, a verifiable delay before the seed, era draws from a genesis reserveMeasured: hourly program swaps on Apple, NVIDIA and AMD cards, 4 October 2026
The mining card does paid, useful, verifiable workPrimecoin's prime chains in 2013 were not useful. Aleo ran proving as consensus and the fastest prover won (both approximate)Proving kept apart from the lottery; shards assigned by sortition, not by speedImplemented: proving v0 and v1 on the devnet from block zero, on NVIDIA cards. The job market for other chains is Designed
A proof-of-work chain where every block is provenProven-execution EVM chains run as rollups on proof-of-stake Ethereum. Conflux has run GPU-mined EVM apps on a DAG since 2020, without proofs (approximate)Proven state on a proof-of-work base layer, produced by the miners themselvesImplemented in part: shards proven and paid on the devnet. Proof verification enforced in consensus is Open
Finality held by miners and not moved by hour-long rentalsDecred votes with stake. Horizen penalises hidden chains. Kaspa limits merge depth (approximate)Vote weight is 30 days of blocks per key. Hashrate that appeared today has no voteImplemented: rule v3 live on the devnet from block zero. External review is owed at gate 3
100% of emission to the people running the hardwareKaspa's fair launch. Zcash and Decred fund developers from emission (approximate)No fee to any team, foundation or fund in the protocol. The miner software's optional 1% dev fee is the one payment to the project, off with one flagImplemented in consensus: the 80/20 coinbase on the devnet
A chain your browser verifies by itselfLight clients trust a committee, as Ethereum's trust a sync committee (approximate)At launch, one execution proof plus a certificate the client is given. The consensus proof that makes the checkpoint self-verifying is phase twoDesigned. The home page's card verifies a devnet certificate in the browser today, with the voter list taken from a node
- - - - - - - + + + + + + +
LayerStateSince
Mining lotteryClass v4 (sub-version 3) from the first block: the latency-shadow program, a new program every hour on Apple, NVIDIA and AMD cards, compiled ahead, the era VDF armed, the ladder at rung 0block zero, 7 Oct 2026
BlocksOne a second is the target; the live page reads the rate from the observerblock zero, 7 Oct 2026
DifficultyRule v2, a 600-second reference window, from the first block (the first devnet switched to it by height at DAA 33,000 on 4 Oct 2026)block zero, 7 Oct 2026
FinalityRule v3: a checkpoint every 30 s of chain, locked at two thirds of all 30-day weight, the weight table frozen at the last lock during a pause. First lock 20:02 UK, 7 Oct 2026. No coin is staked. The only thing at stake is 30 days of public work: a vote key's weight is its blue blocks over the window, and equivocation strips it for 30 daysblock zero, 7 Oct 2026
Provingv0 and v1 from the first block: shards are assigned to miners' keys, proven on their cards, aggregated into segment records and carried in blocks; fees calibratedblock zero, 7 Oct 2026
EmberThe one-click miner, version 0.3.22 on channel devnet-3 (7 Oct 2026); the app window still says Igneum Miner4 Oct 2026, first install
WalletIgneum Wallet 0.1.5 on macOS (0.1.1 first shipped 5 Oct 2026)7 Oct 2026
Mining lotteryClass v4 (sub-version 3) from the first block: the latency-shadow program, a new program every hour on Apple, NVIDIA and AMD cards, compiled ahead, the era VDF armed, the ladder at rung 0block zero
BlocksOne a second is the target; the live page reads the rate from the observerblock zero
DifficultyRule v2, a 600-second reference window, from the first blockblock zero
FinalityRule v3: a checkpoint every 30 s of chain, locked at two thirds of all 30-day weight, the weight table frozen at the last lock during a pause. No coin is staked. The only thing at stake is 30 days of public work: a vote key's weight is its blue blocks over the window, and equivocation strips it for 30 daysblock zero
Provingv0 and v1 from the first block: shards are assigned to miners' keys, proven on their NVIDIA cards, aggregated into segment records and carried in blocks; fees calibratedblock zero
EmberThe one-click miner, on the devnet channel4 Oct 2026, first install
WalletIgneum Wallet on macOS5 Oct 2026, first shipped
-

Measured: the 0.3.22 release record (7 October 2026: genesis, first block, the gate lines and the first lock); engineering log, "first hourly program swap on the live devnet", "difficulty rule v2 activated on the live devnet at DAA 33,000", "first finality lock on the live devnet" (4 October 2026, the first devnet). The block rate and the first lock are rows 7 and 10 of the evidence table.

-

Two caveats, stated here before anyone else states them. Consensus does not yet verify a carried proof; it checks the record's statement against native execution and its signature, and the in-consensus verifier switches on when the proven share of blocks reads one. And the devnet is the project's own machines, its rented fleet and a few outside laptops. Nothing here has been reproduced by anyone outside the project yet; the evidence page says so row by row.

+

Source: the facts page carries the network row; the block rate and the lock are rows of the evidence table.

+

Two caveats, stated here before anyone else states them. Consensus does not yet verify a carried proof. Today, under proving v0, every producer verifies off the consensus path, and consensus checks the record's statement against native execution and its signature. Open: proof verification enforced in consensus (verifier_in_consensus, proof_rule_active_from) is the prerequisite of the no-rescue network exercise (Deliverable 5) and of the proving economy being a protocol guarantee. And the devnet is the project's own machines, its rented fleet and a few outside laptops. Nothing here has been reproduced by anyone outside the project yet; the evidence page says so row by row.

-

Mining: a program that never holds still

-

Every GPU chain that promised ASIC resistance shipped a fixed algorithm, and a fixed algorithm gets a chip the moment the prize pays for one. Igneum does not have a fixed algorithm.

+

Mining: commodity GPUs, scored against a chip

+

Every GPU chain that promised ASIC resistance shipped a fixed algorithm, and a fixed algorithm gets a chip the moment the prize pays for one. Igneum does not assume its algorithm keeps chips away. It assumes a specialised chip exists, seeks profit and stays compatible, and it states how far ahead that chip gets.

Each hour the chain derives a seed from a locked checkpoint one epoch back, passes it through a ten-minute verifiable delay so no miner can see which program a seed implies before choosing whether to publish a block, and feeds it to a deterministic generator. The generator emits a random integer program built from what graphics cards are uniquely good at: wide parallel integer maths, shuffles between the 32 lanes of a warp, and dependent reads spread over a multi-gigabyte dataset that changes daily, so the program waits on memory latency, not on maths or bandwidth. Measured: an RTX 5090 hashes at 95 GB/s of useful 4-byte loads against 1,638 GB/s of sequential writes (engineering log, the RTX 5090 entries). The memory footprint and instruction count are fixed and only the maths sequence is random, so no hour favours one vendor's cards and nobody gains by grinding the seed. Miners compile the program once per hour. Anyone running a node, a wallet or an exchange checks a hash on an ordinary CPU in under ten milliseconds by simulating one warp, so nobody needs a GPU except to mine. Measured: 0.61 ms per warp on one Apple M5 Max core for class v2 and 2.1 ms for class v3 (the mixer at x8, 5 October 2026, one core at load average 5.5, worst cold unit 2.15 ms), 3.4x the class v2 verifier; the 10 ms gate leaves 4.8x (4.6x on the worst cold unit); a 2019-class laptop core is not yet measured.

The hash is a lottery, not a general-purpose cryptographic hash. It has to be unpredictable per nonce, free of any shortcut cheaper than honest evaluation, and free of bias a miner can exploit. It does not need preimage or collision resistance. Open: no analysis of the lottery properties exists yet. It is the first job of the external review in phase 1, and until then the hash is a design claim backed by the measurements below.

@@ -458,23 +458,23 @@ body.all .pager{display:none} - +
Every hourA new random programNo, the miner compiles whatever arrives
Every dayA new datasetNo
Every six monthsA new instruction mix and memory pattern drawn by the chain from rules fixed at genesis, and a new family of instructions unlocked from a reserve written at genesis, so the program space widens every era. A schedule change against fixed datapaths and human forks, not a surprise: a programmable chip reads every drawn parameter as firmwareNo
ContinuouslyThe dataset grows on a schedule fixed at genesis, slowly enough that consumer cards keep up for years. A chip is built with fixed memory, so it is on a countdown from the day it ships. Ethereum's growing dataset ran Bitmain's E3 out of memory in 2020 this way, approximate, with nobody doing anythingNo
ContinuouslyThe dataset grows on a schedule fixed at genesis, slowly enough that consumer cards keep up for years. Each step is scored against the burden it puts on ordinary cards (Deliverable 3); growth is not counted on to retire a chipNo
-

Three ideas carry the chip resistance. The hash rewrites itself. A new program every hour, drawn from the chain. Its memory pattern changes with it. The rules change on a schedule fixed at launch. No release, no vote. These are automatic schedule changes: they defeat a chip wired for one datapath and they need no human fork. Against a chip that stores the dataset every drawn parameter is firmware, and what meets that chip is the latency-shadow work (class v4) and the price per joule (the Horizon lane analysis, 6 October 2026, section 5.4; ledger M32). It waits on memory, not maths. Every hash is a chain of random reads into a table too big for a chip to carry. Measured (8 October 2026; lane D’s family harness at the acceptance rule’s own 2^20 sample over 4,900 drawn eras, and the chained-cache pass’s reading of the night before): every hash’s 128 dependent reads land across the whole dataset and the distinct-index floor holds at 0.995 on every accepted program; about half of epochs carry one load site whose address bit at the era’s stride rotation is biased, which prices about 1.6 percent of a hash’s reads to a chip storing half the dataset and nothing to a chip storing all of it; the next class folds the product’s low bits before the rotation, so no era lands a biased bit on an address bit. The wait is the same physics for everyone. Miners hold the switch. Spare defences are written into the rules, switched off. A miner signal turns one on, at the class-change threshold: miners signal three things at three thresholds, 60 percent of blue blocks over two weeks for a parameter genesis leaves open, 90 percent for an upgrade (new code), and 95 percent with a floor height for a class change. No fork.

+

What carries the chip resistance is the scoring rule's result on the placed adversary rows, plus the economics, reported separately in the chip model below. Three properties feed it. Rotation is an option, not the defence. A new program every hour, drawn from the chain, its memory pattern with it, and rules that change on a schedule fixed at launch. These schedule changes defeat a chip wired for one datapath and need no human fork, but against a programmable chip that stores the dataset every drawn parameter is firmware, so the security argument does not rest on them. What meets that chip is the latency-shadow work (class v4) and the price per joule (the chip and economy analysis of 6 October 2026, section 5.4; ledger M32). It waits on memory, not maths. Every hash is a chain of random reads into a table too big for a chip to carry. Measured (8 October 2026; lane D’s family harness at the acceptance rule’s own 2^20 sample over 4,900 drawn eras, and the chained-cache pass’s reading of the night before): every hash’s 128 dependent reads land across the whole dataset and the distinct-index floor holds at 0.995 on every accepted program; about half of epochs carry one load site whose address bit at the era’s stride rotation is biased, which prices about 1.6 percent of a hash’s reads to a chip storing half the dataset and nothing to a chip storing all of it; the next class folds the product’s low bits before the rotation, so no era lands a biased bit on an address bit. The wait is the same physics for everyone. Miners hold the switch. Spare defences are written into the rules, switched off. A miner signal turns one on, at the class-change threshold: miners signal three things at three thresholds, 60 percent of blue blocks over two weeks for a parameter genesis leaves open, 90 percent for an upgrade (new code), and 95 percent with a floor height for a class change. No fork.

The work that waits can grow. Class v4 adds a block of latency-shadow arithmetic to every hash, about 100,000 integer operations that run while the memory reads are in flight, so a chip that stores the whole dataset still has to pay for a core. That size sits on a ladder fixed at genesis, six rungs from about 100,000 to about 1,000,000 operations, and it moves one rung at a time only when 90 percent of blue blocks in each of seven consecutive days ask for it; it can never move two rungs inside a week and never past a rung the reference verifier cannot check under 10 ms with its sibling thread busy (measured on the build server, 6 October 2026: the first three rungs pass at 8.8, 8.9 and 9.2 ms, the fourth misses by 0.08 ms on a loaded box and stays out until a quiet re-measurement, the two doublings are out at 12.4 and 15.0 ms). What it buys against a chip is scored under the rule in the chip model below. What it costs, per rung, is measured too: the Apple tier gives up 3 points of rate at the first step and 6 more at the second, the RTX 5090 nothing until the second; so the miners who pay for a step are the ones who take it (ledger M34).

The chip model

Igneum remains competitive on accessible commodity GPUs even when specialised mining hardware is assumed to exist, remain compatible and seek profit; its security does not rely on identifying that hardware or retiring it through emergency changes. Class v6 adopts the 64-register window and retains it across every rotation. Current modelling places the strongest specialised designs assessed against the GPU tier at about 2.3x to 3.3x energy-efficiency advantage a node ahead (2.0x to 2.9x on the GPU's own node), a bracket that is approximate and provisional until the placed gated core rows land. The long-program and select-tree proposals were rejected. Economic resistance depends on development cost, deployment economics and productive hardware lifetime; family transitions receive an obsolescence benefit only where a loss of competitiveness is demonstrated; programmable multi-epoch designs are included in the assessment.

The labels. The bracket is modelled and provisional: its floor is the clock-gated base core and its ceiling the first placed core, which came in 64 percent above synthesis (wires and the clock tree); the honest figure is the placed gated core’s and replaces the bracket when its row lands. The GPU side is measured: an RTX 5080 at its 1,100 MHz core lock, 2.06 microjoules per hash, and an RTX 5090 at its 1,300 MHz lock, 2.33 microjoules per hash, both under class v4, on the project’s own rigs and rented pods, 8 October 2026; the card’s cost of the window is measured too (a rented RTX 5090 and RTX 4090 at stock, 8 October 2026: within 5 percent per load with the liveness chain, no register spill). The chip side is synthesised and claimed: the clock-gated sequencer core with the 64-register window on ASAP7, scaled to N3 on the foundry’s headline factors (k about 0.37 at N3 and 0.51 node for node for the base core, the gated window adding about 0.13 of k against an adversary with a flop register file; the window’s liveness measured at 61 of 64 values necessary, its cost to the card measured under 5 percent); the window’s k is synthesis-derived and not a lower bound, and the multi-family adversary lane’s first core (its state in a macro) reads the window’s defence as close to nothing, a disagreement between two models that the placed rows settle. The chip’s memory is modelled: the GDDR7 board of the chip model. The placed gated figure is expected near 2.6x to 3.1x a node ahead and 2.3x to 2.7x node for node (approximate) and is served when its row lands.

Three statements, kept separate. The baseline is the hash as it stands under the scoring rule; rotation is an optional improvement to that baseline, not the mechanism the claim rests on.

- - - + + +
StatementWhat it saysLabel and date
Energy resistanceThe bracket above: about 2.3x to 3.3x for the strongest specialised design a node ahead of the GPU tier, 2.0x to 2.9x on the GPU’s own node, provisional until the placed gated core row lands; two nodes ahead follows from that row. The honest tier moves to the next node with every GPU generation; a chip must tape out again.modelled on measured cards, 8 October 2026, approximate and provisional; the node column is claimed scaling
Economic resistanceWhether a chip gets built depends on development cost, deployment economics and productive hardware lifetime. The first cut of the profitability surface: the price at which a project pays scales as the project cost over its share of the chain times its discounted life, and moves by under 5 percent with the per-joule edge; a fixed-lane chip under rotation needs 4x the price a programmable one needs. No threshold is the headline: the coexistence model that prices the conditions (docs/analysis/class-v6/coexistence-model.md) is owed and is served when it exists.modelled, first cut, 8 October 2026; conditional until the cut lands
Response capabilityRotation is an optional improvement, not the mechanism. A passed rotation boundary proves the rotation works, not that hardware dies. The schedule: a new program every hour, a parameter era every week, a family epoch every 180 days, an emergency vote when miners call one.measured per boundary, 8 October 2026
Energy resistanceThe bracket above: about 2.3x to 3.3x for the strongest specialised design a node ahead of the GPU tier, 2.0x to 2.9x on the GPU’s own node, provisional until the placed gated core row lands; two nodes ahead follows from that row. The honest tier moves to the next node with every GPU generation; a chip must tape out again. Whole machine per tier (synthesis with the SRAM band and node factors, claimed; placed rows to follow): the complete GDDR7 machine about 1.8x the RTX 5090 at its lock per joule node for node and 2.1x a node ahead; 1.6x and 1.9x the RTX 5080; 2.8x and 3.3x the Ada, Ampere and RX 9070 XT cohort; about 1.5x the Apple tier, reported, never headlined.modelled on measured cards, 8 October 2026, approximate and provisional; the node column is claimed scaling; the per-tier line claimed, 8 October 2026
Economic resistanceWhether a chip gets built depends on development cost, deployment economics and productive hardware lifetime. The first cut of the profitability surface: the price at which a project pays scales as the project cost over its share of the chain times its discounted life, and moves by under 5 percent with the per-joule edge; a fixed-lane chip under rotation needs 4x the price a programmable one needs. No threshold is the headline: the five-year coexistence model (Deliverable 4; its first run is docs/analysis/class-v6/coexistence-model.md, every row modelled) replaces any capex wall: the GDDR7 board passes all six of its success conditions at a one to three year life; an N2 SRAM die fails five once it exists with development sunk, and the only condition holding it is that nobody pays to build it; the larger half of a chip's edge is capital cost per accepted hash, not joules.modelled, the coexistence model's first run, 8 October 2026
Response capabilityRotation is an optional improvement, not the mechanism. A passed rotation boundary proves the rotation works, not that hardware dies. The schedule: a new program every hour, a parameter era every week, a family epoch every 180 days, an emergency vote when miners call one. Rotation costs a chip versatility, not life: the family bank is firmware plus about 43 percent of core cells, and no transition carries an obsolescence credit (modelled).measured per boundary, 8 October 2026; the family-bank cost modelled, 8 October 2026
-

What a miner sees from this. Class v4 costs a 5090 145 W more unlocked, 88 W more at a 1,400 MHz core lock and 82 W at the best operating points (class v4 at 1,200 MHz, class v3 at 1,300; the knee is 1,300 MHz on both), for 0.2 percent more rate (measured, 7 October 2026; the 80 W read on 6 October was at the app's tuned cap); an M5 Max 16 W more for 1.5 percent, an RX 9070 XT and an RTX 4070 nothing (measured, 6 October 2026). The ladder that sets how much work rides in the shadow starts at rung 0 at genesis and climbs by miner signal; its third rung is inadmissible today because a server core verifies it in 10.85 ms, over the gate (measured, 7 October 2026). Devnet 3 runs class v4 from its first block (7 October 2026); the first devnet started on class v3 and reaches class v4 by miner signal at a published height. Classes rotate on findings and at least yearly; a class change is a release activated by block height. Class v5 crosses on Devnet 3 by height; class v6 is the design in progress (opened 8 October 2026), with four layers as its spine: per-era draws of the parameters a release now fixes, a dataset whose size tracks the chain state, scheduled family epochs by height, and the acceptance floor generalised to every era’s draw. The next test of the model is an internal adversarial pass, not an independent review: three lanes that have never worked on the hash code attack the mixer, the chained cache and the acceptance rule with only what an outsider has (the public kit, the frozen object, the spec, the harnesses) and publish the break or the bound they reach. No outside review has run yet.

-

No hash has stayed free of chips forever. Igneum does not claim to. It states the gain its own model finds and the response the rotation makes, and both carry their labels. The precedents, as sourced (nameplate and community tables, about 20 percent either way; every figure with its URL and date in the close): Monero has run on RandomX since November 2019, its rules stable since then and its programs varying per hash; one chip shipped against it, Bitmain’s Antminer X5 (September 2023), 46 months after the fork, at 6.37 J per kH at the wall against a stated CPU measurement, an observed comparison, not a ceiling. Bitmain opened Antminer X9 pre-orders on 26 December 2025 for July 2026 delivery, then withdrew the product in mid-May 2026 and refunded buyers before any unit shipped; none has been independently benchmarked. RandomX v2 was released on 25 March 2026 with its mainnet activation pending. Ethash ran 36 months to a first chip worse than a GPU; the iPollo V2H reads about 14x today. Kaspa ran 21 months to its first chip, at 167x to 725x. The commodity cohort Igneum protects is the discrete-GPU population; the Apple row is reported beside it, never as the headline.

-

The scoring rule and the harness. The edge is the minimum over workloads of the maximum over free adversarial designs of the GPU’s joules per hash over the adversary’s, under four conditions: the 10 percent GPU-cost budget at the lock, the verifier limit, cross-vendor correctness and hardware accessibility. The rejected designs stand as negative controls with their measured rows: the long program, the select tree, the wide read, the scratchpad. The next programme: the connected-state experiment, the mixed integer and FP32 candidate, the multi-family programmable adversary; rotation is not expected to deliver the missing joule. The scoring rules and every row, section 10. The harness and its readings (measured, 8 October 2026): the acceptance floor at 0.995 refuses nine of nine hot sets (0.9809 to 0.9919) and 2.435 percent of 4,600 drawn programs, 0 of 61 in the era reading (section 14); the attack families F8, F4, F9 and F1 pass, F8 with a residue of 61 of 64 (section 13); the attempts census and the attempt-3 read (section 0).

+

What a miner sees from this. Class v4 costs a 5090 145 W more unlocked, 88 W more at a 1,400 MHz core lock and 82 W at the best operating points (class v4 at 1,200 MHz, class v3 at 1,300; the knee is 1,300 MHz on both), for 0.2 percent more rate (measured, 7 October 2026; the 80 W read on 6 October was at the app's tuned cap); an M5 Max 16 W more for 1.5 percent, an RX 9070 XT and an RTX 4070 nothing (measured, 6 October 2026). The ladder that sets how much work rides in the shadow starts at rung 0 at genesis and climbs by miner signal; its third rung is inadmissible today because a server core verifies it in 10.85 ms, over the gate (measured, 7 October 2026). The devnet runs class v4 from its first block. Classes rotate on findings and at least yearly; a class change is a release activated by block height. Class v5 is built to cross by height; its dataset keyed by the chain's own state is under evaluation (Deliverable 3) and is not counted as a defence until justified. Class v6 is the design in progress (opened 8 October 2026), with four layers as its spine: per-era draws of the parameters a release now fixes, a dataset whose size tracks the chain state (under evaluation, Deliverable 3), scheduled family epochs by height, and the acceptance floor generalised to every era’s draw. The next test of the model is an internal adversarial pass, not an independent review: three lanes that have never worked on the hash code attack the mixer, the chained cache and the acceptance rule with only what an outsider has (the public kit, the frozen object, the spec, the harnesses) and publish the break or the bound they reach. No outside review has run yet.

+

No hash has stayed free of chips forever. Igneum does not claim to. It states the gain its own model finds, the economics and the response capability, each separately, and each carries its label. The precedents, as sourced (nameplate and community tables, about 20 percent either way; every figure with its URL and date in the close): Monero has run on RandomX since November 2019, its rules stable since then and its programs varying per hash; one chip shipped against it, Bitmain’s Antminer X5 (September 2023), 46 months after the fork, at 6.37 J per kH at the wall against a stated CPU measurement, an observed comparison, not a ceiling. Bitmain opened Antminer X9 pre-orders on 26 December 2025 for July 2026 delivery, then withdrew the product in mid-May 2026 and refunded buyers before any unit shipped; none has been independently benchmarked. RandomX v2 was released on 25 March 2026 with its mainnet activation pending. Ethash ran 36 months to a first chip worse than a GPU; the iPollo V2H reads about 14x today. Kaspa ran 21 months to its first chip, at 167x to 725x. The commodity cohort Igneum protects is the discrete-GPU population; the Apple row is reported beside it, never as the headline.

+

The scoring rule and the harness. The edge is the minimum over workloads of the maximum over free adversarial designs of the GPU’s joules per hash over the adversary’s, under four conditions: the 10 percent GPU-cost budget at the lock, the verifier limit, cross-vendor correctness and hardware accessibility. The rejected designs stand as negative controls with their measured rows: the long program, the select tree, the wide read, the scratchpad. The two architectural experiments are closed as failures and stay as regression controls: the mixed integer and FP32 branch (measured, 8 October 2026: 15 to 26 percent more card energy per hash against the 10 percent budget) and the connected-state reorganisation (8 October 2026: only the window width reaches the chip). The multi-family programmable adversary (Deliverable 3) is open; rotation is not expected to deliver the missing joule. The scoring rules and every row, section 10. The harness and its readings (measured, 8 October 2026): the acceptance floor at 0.995 refuses nine of nine hot sets (0.9809 to 0.9919) and 2.435 percent of 4,600 drawn programs, 0 of 61 in the era reading (section 14); the attack families F8, F4, F9 and F1 pass, F8 with a residue of 61 of 64 (section 13); the attempts census and the attempt-3 read (section 0).

One thing takes a person, here and on every chain that exists: writing new code. A chain cannot safely write its own generator, and it cannot safely tell a chip from a wave of honest new cards by hashrate alone. If the design above ever failed, anyone could publish a new generator and miners would switch it on by signalling, as Monero's community can fork. Igneum is built to make that day unlikely, and does not depend on avoiding it.

@@ -486,27 +486,41 @@ body.all .pager{display:none} Hardware it is built forCPUs. GPUs run it badly on purposeGPUs. Any card, any vendor. Bit-exact on Apple, NVIDIA and AMD, measured Random programPer hash, interpreted in a virtual machinePer hour, compiled to native GPU code. Per hash, the 128 dataset addresses change with the nonce - DatasetAbout 2 GB, the same size since 2019, approximate2 GB, growing (the proposed schedule, fixed at the testnet genesis: 2 GB, doubling at years 4, 12 and 28); a 4 GB card mines about four years, an 8 GB card about twelve, approximate + DatasetAbout 2 GB, the same size since 2019, approximate2 GB, growing (the proposed schedule, fixed at genesis: 2 GB, doubling at years 4, 12 and 28); a 4 GB card mines about four years, an 8 GB card about twelve, approximate Light verification256 MB cache on a CPU, milliseconds256 MB cache on a CPU (512 MB from year 4), one warp under 10 ms, the gate. Measured 2.1 ms on one Apple M5 Max core for class v3 (3.4x class v2's 0.61 ms); a 2019-class core not yet Changes over timeNone. A fixed design, unchanged for seven yearsA new program every hour, its memory pattern with it; era draws and reserved families on a schedule fixed at genesis. Nobody touches it Seed grindingNot applicable, the program comes from the hash inputClosed by a verifiable delay between seed and program - Useful workNone. Hashing onlyEvery NVIDIA card from 8 GB proves; 12 GB and up mine and prove; 24 GB on the stock server (eleven rented cards, RTX 3060 to RTX 5090, 6 October 2026); they sell proofs to other chains. AMD and Apple cards mine, and a prover for them lands when a zkVM ships one + Useful workNone. Hashing onlyNVIDIA cards prove: from 8 GB on the patched server, 12 GB and up beside the miner, 24 GB on the stock server (measured on eleven rented cards, RTX 3060 to RTX 5090, 6 October 2026). Selling proofs to other chains is Designed, not built. AMD and Apple cards mine and do not prove; a prover for them lands when a zkVM ships one Track recordAbout seven years with one shipped chip, Bitmain’s Antminer X5 (September 2023), at 1.46x per joule over a desktop CPU; the one announced beyond it, Bitmain’s Antminer X9, was withdrawn in May 2026 with zero units; RandomX v2 released 25 March 2026, activation pendingZero years. Every number above is measured and logged with the commands that produced it. The specification, reference hash, test vectors and simulators are public now (git.igneum.network/igneum-network/spec). The node, the miner and the wallet follow to the same host as the repository is published -

Measured so far: the same hourly program, generated on an Apple M5 Max, compiled by Apple's Metal and NVIDIA's CUDA on an RTX 5090, produced identical hashes on both, 192 of 192 across two programs. On a 1 GB dataset the 5090 ran at about 228 million hashes a second and the Mac at about 45 million, both bound by random memory access rather than arithmetic. Those are prototype figures, not mining rates. The first prototype dataset was a closed-form function, and a miner could compute items instead of loading them: measured 111x faster that way on the Mac. The 256 MB cache construction replaced it on 3 October 2026. With the cache, computing items on the fly runs 4.8x slower than loading them, measured on the Mac, and the honest rate is unchanged on both vendors. Open: the same shortcut ratio on NVIDIA and on a discrete AMD card, and the time-memory trade-off between the two measured points. Inside the 5090's 96 MB cache the same program ran nearly six times faster, which is why the dataset starts at 2 GB and grows. On 4 October 2026 the live devnet crossed an hourly program change on all three vendors with no pause and no rejected block: a Mac at 26.7 million hashes a second, an RTX 5090 at 123 million and an integrated AMD chip at 2.7 million, every hash doing 128 distinct reads of the memory-hard dataset.

+

Measured so far: the same hourly program, generated on an Apple M5 Max, compiled by Apple's Metal and NVIDIA's CUDA on an RTX 5090, produced identical hashes on both, 192 of 192 across two programs. On a 1 GB dataset the 5090 ran at about 228 million hashes a second and the Mac at about 45 million, both bound by random memory access rather than arithmetic. Those are prototype figures, not mining rates. The first prototype dataset was a closed-form function, and a miner could compute items instead of loading them: measured 111x faster that way on the Mac. The 256 MB cache construction replaced it on 3 October 2026. With the cache, computing items on the fly runs 4.8x slower than loading them, measured on the Mac, and the honest rate is unchanged on both vendors. Open: the same shortcut ratio on NVIDIA and on a discrete AMD card, and the time-memory trade-off between the two measured points. Inside the 5090's 96 MB cache the same program ran nearly six times faster, which is why the dataset starts at 2 GB and grows. On 4 October 2026 a live network crossed an hourly program change on all three vendors with no pause and no rejected block: a Mac at 26.7 million hashes a second, an RTX 5090 at 123 million and an integrated AMD chip at 2.7 million, every hash doing 128 distinct reads of the memory-hard dataset.

Proving: the miners are the provers

-

Every Igneum block is proven with a zero-knowledge proof, and the miners produce it. Proving is a useful GPU workload that is cheaply verifiable by construction. A proof is right or it is not. Wrapped for light clients, a phone checks it in milliseconds; the wrapping cost is a phase two measurement. Measured so far, the certificate half only: the browser verifier on the home page checks a devnet finality certificate, one BLS aggregate signature over 16 keys and 21 header hashes, in 139 to 155 ms cold and 58 to 68 ms warm in a phone-sized tab on a laptop core (5 October 2026). No phone has been measured, and no wrapped block proof exists yet.

+

A GPU-secured network for Ethereum-compatible applications and verifiable computation.

+

The proof architecture

+

Igneum is not an Ethereum L2. Ethereum does not enforce its state or hold its data, so Igneum carries its own consensus security, data availability and cross-chain verification. The architecture is three decisions, kept separate, and one source of extra demand. EVM-compatible execution is what developers build against, through revm. SP1 is the one well-tested proving backend, behind the versioned proving interface; a zkEVM here means the EVM implementation compiled as a program SP1 proves. Igneum's own GPU-mined consensus is where security comes from. External customers are the source of additional proving demand: designed, not built, and out of every revenue assumption.

+
+ + + + + + + +
DecisionChoiceState, 8 Oct 2026
What developers build againstEVM-compatible execution (revm), with a documented set of differences: block context, randomness, two-dimensional feesImplemented on the devnet
How execution is provedSP1, one well-tested backend behind the versioned proving interface; program identities, verifier versions and security parameters pinned in the protocol; a second backend only where justifiedImplemented: proving v0 and v1 on the devnet. The pinning is Designed
Where security comes fromIgneum's own GPU-mined consensus: the lottery, GHOSTDAG ordering and miner-only finalityImplemented on the devnet; external review owed
Additional proving demandExternal customers buying proofs for their own systemsDesigned, not built; out of revenue assumptions
+

Proven execution is not finality. EVM compatibility is not Ethereum security. ZK is not privacy.

+

A proof says the state follows from the ordered blocks; the miners' lock decides which blocks are final. Running Ethereum's bytecode does not bring Ethereum's validators. Published state data is public, and privacy needs its own application or protocol design. None of these choices, by itself, answers specialised mining hardware; the chip model is a separate obligation.

+

Igneum blocks are proven with zero-knowledge proofs, and NVIDIA miners produce them. AMD and Apple cards mine; they do not prove today. Proving is a useful GPU workload that is cheaply verifiable by construction. A proof is right or it is not. Wrapped for light clients, a phone checks it in milliseconds; the wrapping cost is a phase two measurement. Measured so far, the certificate half only: the browser verifier on the home page checks a devnet finality certificate, one BLS aggregate signature over 16 keys and 21 header hashes, in 139 to 155 ms cold and 58 to 68 ms warm in a phone-sized tab on a laptop core (5 October 2026). No phone has been measured, and no wrapped block proof exists yet.

How a block gets proven

-

Blocks carry transactions only and make no claim about state. Every node executes the ordered transactions natively at once, so users see their transaction land in about a second. The execution is then split into shards of a fixed proving cost. Shards are assigned by lot to eight provers for ten seconds, then open to anyone; there is no bond. Provers run them on consumer cards, and the shard proofs are folded by recursive aggregation into one proof for the block. That proof lands on-chain within about a minute at launch. Because the proof computes the state from the ordered sequence, no node accepts a block with a wrong state root. Full nodes also execute every block natively and reject a proof record whose result differs from their own execution, so a forged proof is a light-client problem and never a chain split. Implemented: the native-execution check on every carried proof record, proving v0 on the devnet (specification section 7). The emergency path for a soundness bug in the proof system is a human one: a new proof-system version is written by people and activates only on miner signalling. Invalid transactions are skipped by rule, the way Kaspa skips conflicting spends.

-

Proving: every NVIDIA card from 8 GB proves; 12 GB and up mine and prove; 24 GB on the stock server. Measured on eleven rented cards, RTX 3060 to RTX 5090, 6 October 2026: the RTX 3060 (12 GB) mines at 23.78 MH/s and proves the v1 shard beside its miner at an 8.9 GB peak in 37.5 s; the RTX 4060 (8 GB) proves it alone at 7.4 GB in 18.4 s; the RTX 4090 (24 GB) proves it on the stock SP1 server in 5.6 s at 17.4 GB. The patched server that fits the smaller cards is not yet in the shipped app. AMD and Apple cards mine. A prover for them lands when a zkVM ships one.

+

Blocks carry transactions only and make no claim about state. Every node executes the ordered transactions natively at once, so users see their transaction land in about a second. The execution is then split into shards of a fixed proving cost. Shards are assigned by lot to eight provers for ten seconds, then open to anyone; there is no bond. Provers run them on consumer NVIDIA cards, and the shard proofs are folded by recursive aggregation into one proof for the block. That proof lands on-chain within about a minute at launch (designed). Because the proof computes the state from the ordered sequence, no node accepts a block with a wrong state root. Full nodes also execute every block natively and reject a proof record whose result differs from their own execution, so a forged proof is a light-client problem and never a chain split. Implemented: the native-execution check on every carried proof record, proving v0 on the devnet (specification section 7). The emergency path for a soundness bug in the proof system is a human one: a new proof-system version is written by people and activates only on miner signalling. Invalid transactions are skipped by rule, the way Kaspa skips conflicting spends.

+

Proving is NVIDIA's today: from 8 GB on the patched server, 12 GB and up beside the miner, 24 GB on the stock server. Measured on eleven rented cards, RTX 3060 to RTX 5090, 6 October 2026: the RTX 3060 (12 GB) mines at 23.78 MH/s and proves the v1 shard beside its miner at an 8.9 GB peak in 37.5 s; the RTX 4060 (8 GB) proves it alone at 7.4 GB in 18.4 s; the RTX 4090 (24 GB) proves it on the stock SP1 server in 5.6 s at 17.4 GB. The patched server that fits the smaller cards is not yet in the shipped app. AMD and Apple cards mine. A prover for them lands when a zkVM ships one. These rows are the proving stage only: the full pipeline is judged, inputs, proving, aggregation, verification, payment, memory and the mining income forgone, and a fast shard does not settle it.

The proving budget

-

Gas prices execution. Proving cost is a different number, so Igneum meters it separately: every transaction pays in both dimensions, and each block has a proving-cost budget set in consensus from measured prover throughput. A transaction that is cheap to run and expensive to prove pays for what it costs the provers. Measured on 5 October 2026 (an RTX 5090 under SP1 6.8.1's GPU prover, the shard size the chain adopts from its fee switch, 30,000 proving gas, about 4.7 million prover cycles): one full shard proves in 4.3 seconds and needs 20.4 GB of GPU memory with the card to itself, so a 24 GB card proves full shards and a 12 GB or 16 GB card does not on this prover build, whose floor is 13.9 GB for even an empty shard; mining and proving on one card needs 32 GB today (the prototype-size shard beside the miner peaked at 30.1 GB) and 24 GB once the adopted shard size is live (22.2 GB beside the miner, 13.2 seconds a shard, measured on the 32 GB card; a 24 GB card has not run it yet). The old 12 GB gate on the roadmap was withdrawn on 5 October until a prover build with a smaller floor was measured; on 6 October a patched server proved the same shard at 7.4 to 8.0 GB alone on eleven rented cards from the RTX 3060 to the RTX 5090 (the real-card table), so the gate returns as measured and the patched server is not yet in the shipped app. The first proofs exist: on 4 October 2026 an RTX 5090 proved a small two-transaction block in 1.4 seconds (2.7 seconds compressed), verified in 0.22 and 0.038 seconds, and a laptop CPU proved a three-shard block end to end in 19 minutes. Later that day the same card proved a full shard at the provisional size, 6.75 million prover gas, which executed in 60.8 million cycles: core proof 8.3 seconds, compressed proof 10.9 seconds, verified in 0.040 seconds; a four-shard block took 44.5 seconds of GPU stages end to end. Since 5 October 2026 shards are assigned and proven on the live devnet. The gate asks for a mid-range card, and an RTX 5090 is not one, so the gate stands open. Once the gate is measured, the budget rises by schedule as hardware improves. The proof system is hash-based, which is what runs on consumer cards, and sits behind a versioned interface, so Igneum can adopt a better proof system when one exists by a miner-signalled release, and runs for ever on the current one if none is adopted.

+

Gas prices execution. Proving cost is a different number, so Igneum meters it separately: every transaction pays in both dimensions, and each block has a proving-cost budget set in consensus from measured prover throughput. A transaction that is cheap to run and expensive to prove pays for what it costs the provers. Measured on 5 October 2026 (an RTX 5090 under SP1 6.8.1's GPU prover, the shard size the chain adopts from its fee switch, 30,000 proving gas, about 4.7 million prover cycles): one full shard proves in 4.3 seconds and needs 20.4 GB of GPU memory with the card to itself, so a 24 GB card proves full shards and a 12 GB or 16 GB card does not on this prover build, whose floor is 13.9 GB for even an empty shard; mining and proving on one card needs 32 GB today (the prototype-size shard beside the miner peaked at 30.1 GB) and 24 GB once the adopted shard size is live (22.2 GB beside the miner, 13.2 seconds a shard, measured on the 32 GB card; a 24 GB card has not run it yet). The old 12 GB gate on the roadmap was withdrawn on 5 October until a prover build with a smaller floor was measured; on 6 October a patched server proved the same shard at 7.4 to 8.0 GB alone on eleven rented cards from the RTX 3060 to the RTX 5090 (the real-card table), so the gate returns as measured and the patched server is not yet in the shipped app. The first proofs exist: on 4 October 2026 an RTX 5090 proved a small two-transaction block in 1.4 seconds (2.7 seconds compressed), verified in 0.22 and 0.038 seconds, and a laptop CPU proved a three-shard block end to end in 19 minutes. Later that day the same card proved a full shard at the provisional size, 6.75 million prover gas, which executed in 60.8 million cycles: core proof 8.3 seconds, compressed proof 10.9 seconds, verified in 0.040 seconds; a four-shard block took 44.5 seconds of GPU stages end to end. Shards are assigned and proven on the devnet from block zero. The gate asks for a mid-range card, and an RTX 5090 is not one, so the gate stands open. Once the gate is measured, the budget rises by schedule as hardware improves. The proof system is hash-based, which is what runs on consumer cards, and sits behind a versioned interface. SP1 is the one backend. A replacement is adopted only where justified, by a miner-signalled release, never as an interchangeable second backend, and the chain runs for ever on the current one if none is adopted.

Proving for everyone else

-

The same miners accept proving jobs from other chains. Rollups post a job, a miner wins it, proves it, and is paid. The job market is permissionless and is Designed, not yet built. At launch a job is paid on the customer's own chain, in the customer's currency, to a payout contract keyed by miner address, because Igneum cannot yet see Ethereum. Settlement in IGN, with 10% of each fee burned, follows when the proof bridge lets Igneum see the payment, in phase two. The Igneum miner client can also bid on other proving networks and take the best price, where a miner chooses to hold their collateral: Boundless provers post ZKC and Succinct provers stake PROVE (approximate, from their documentation). The proving market is small today. Igneum does not depend on it. We know of no other proof-of-work chain selling proofs to other chains. Live rows arrive with the public testnet. The public testnet is armed: three seed nodes and the public RPC are up, and it opens on the go word.

+

The job market for other chains is Designed, not built, and stays out of every revenue assumption until it is. The order: Igneum's own execution first; then one external customer's exact workload with repeat paid jobs; further workloads only where the fleet has a demonstrated edge. As designed, a customer posts a job, a miner wins it, proves it, and is paid, and the market is permissionless. At launch a job is paid on the customer's own chain, in the customer's currency, to a payout contract keyed by miner address, because Igneum cannot yet see Ethereum. Settlement in IGN, with 10% of each fee burned, follows when the proof bridge lets Igneum see the payment, in phase two. The Igneum miner client can also bid on other proving networks and take the best price, where a miner chooses to hold their collateral: Boundless provers post ZKC and Succinct provers stake PROVE (approximate, from their documentation). The proving market is small today. Igneum does not depend on it. We know of no proof-of-work chain that sells proofs to other chains.

@@ -516,7 +530,7 @@ body.all .pager{display:none}

Finality

Every 30 seconds of chain a checkpoint forms, deep enough past the tip that the DAG will not reorder it. Every miner with at least 100 blocks in the last 30 days signs it, and the checkpoint locks when signatures representing two thirds of all the mining weight of those 30 days arrive. Once a checkpoint is locked it overrides the heaviest chain, so fresh hashrate cannot reorganise past it. Two thirds of 30-day weight can. In the chain's first 30 days no checkpoint locks at all: the rule waits until the window holds 30 days of history (Implemented, the first-month gate, measured on test networks on 4 and 5 October 2026), so the chain runs on proof of work and its 12-hour finality depth the way every new proof-of-work chain does. On the devnet, whose window is two hours, the first lock came two hours after genesis, at 77.4% of all weight from 17 vote keys (4 October 2026).

The word sustained is the whole defence. Block rewards go to whoever mines, new or old. The right to lock history is earned.
-

A miner's vote weight is simply the blocks it has mined over the trailing 30 days, measured by work, so splitting into many keys buys nothing and joining a pool costs nothing. Hashrate that arrived today holds almost none of it. Even an attacker producing every block on the chain, with honest miners gone, would need ten days of mining in public to hold a third of the weight, and twenty to hold two thirds. An attacker matching the honest network needs twenty days for a third and never reaches two thirds while the honest miners keep mining. Rental is priced by the hour. The only route left is to drive honest miners off the chain and hold two thirds for a month on the public hashrate charts, which is the same limit Bitcoin lives with, with a month's warning attached. Pools carry their hashers' votes, so vote concentration equals pool concentration, and it is public.

+

A miner's vote weight is simply the blocks it has mined over the trailing 30 days, measured by work, so splitting into many keys buys nothing and joining a pool costs nothing. Hashrate that arrived today holds almost none of it. Even an attacker producing every block on the chain, with honest miners gone, would need ten days of mining in public to hold a third of the weight, and twenty to hold two thirds. An attacker matching the honest network needs twenty days for a third and never reaches two thirds while the honest miners keep mining. Rental is priced by the hour. The only route left is to drive honest miners off the chain and hold two thirds for a month on the public hashrate charts, which is the same limit Bitcoin lives with, with a month's warning attached. Pools carry their hashers' votes, so vote concentration equals pool concentration, and it is public (today's behaviour, as built). The pin replaces it: vote keys stay with the miner at protocol level: the member's retained voting key is committed into its work, payment aggregation is separate and verifiable, and pool identity substitution is resisted (a pin of Igneum 2.0, pools and participation; not yet shipped).

Two further rules close the gaps. A lock needs two thirds of all 30-day weight, so finality pauses whenever less than two thirds of that weight is connected and signing, until it returns or ages out of the window, up to 30 days, and the chain runs on proof of work meanwhile. The node reports the pause. A key that stops signing is reported as absent within two hours, which is how operators see a pause coming. Beneath the latest lock the depth to rely on is the finality depth: a node never switches to a chain forked more than 12 hours of median time back, and a certified checkpoint shortens that to its own age. Kaspa's one-hour merge depth is a limit on which old blocks a new block may merge, not a reorganisation bound. Signing two different checkpoints at the same height is equivocation, provable by anyone, and it strips the key of its vote for 30 days.

What is not here

No coin is staked. The only thing at stake is 30 days of public work: a vote key's weight is its blue blocks over the window, and equivocation strips it for 30 days. No coin-holder class votes on anything. No anchoring into Bitcoin or any other chain. Nothing in Igneum's consensus depends on anything outside Igneum.

@@ -524,7 +538,8 @@ body.all .pager{display:none}

Building on Igneum

-

Anything that runs on Ethereum runs on Igneum unchanged. Same Solidity, same bytecode, same wallets, same tools, a different chain id. Builders get Ethereum semantics with one-second inclusion, finality in about two minutes, and gas priced for a chain that is not congested.

+

A GPU-secured network for Ethereum-compatible applications and verifiable computation.

+

Contracts built for Ethereum deploy with the same Solidity, the same bytecode, the same wallets and tools, and a different chain id. Compatibility is shown, not assumed: representative contracts, wallet fee estimation, indexing, failed transactions and receipts are tested as a product deliverable, and the differences are documented (block context, randomness, two-dimensional fees). Builders get Ethereum semantics with one-second inclusion, finality in about two minutes, and gas priced for a chain that is not congested.

Three things Igneum offers at the base layer that we know no other EVM chain offers.

  1. Proving as a native primitive. A contract can request a proof of any computation and pay for it in gas, and the miners produce it. A game proves a fair shuffle. A lending market proves its solvency. A rollup elsewhere posts a job and gets its proof back. We know of no other EVM chain with a prover network in its base layer.
    2. @@ -535,7 +550,7 @@ body.all .pager{display:none}

    Why build here

    Not for speed. Fast EVM chains filled with copied Ethereum contracts and emptied when incentives stopped. Three things no L2 can offer. Keep your Ethereum deployment.

      -
    1. Proofs priced by the subsidy forgone. A contract requests a proof of any computation and the miners produce it. Their cards already run and are paid by emission, so a job has to beat the lottery income the card forgoes while it proves. That is the price a prover must charge, as a formula with network hash as the input: per shard, (card hash ÷ network hash) × 0.8 × 31.688 IGN × shard seconds, plus electricity, which is under a cent per billion cycles on every card. It falls as one over network hash: at the devnet's 1.16 GH/s a quote is 100 to 300x the published market rate; a card proving beside its miner is competitive near 100 GH/s (the Horizon economy lane, 6 October 2026, sections 3.1 and 4.1, approximate beyond the one card measured; ledger E20). Verification is folded into the chain's own proof; you ship no verifier. The job's base fee rises with the backlog, published at the phase 4 job market.
      2. +
    2. Proofs priced by the subsidy forgone. A contract requests a proof of any computation and the miners produce it. Their cards already run and are paid by emission, so a job has to beat the lottery income the card forgoes while it proves. That is the price a prover must charge, as a formula with network hash as the input: per shard, (card hash ÷ network hash) × 0.8 × 31.688 IGN × shard seconds, plus electricity, which is under a cent per billion cycles on every card. It falls as one over network hash: at the devnet's 1.16 GH/s a quote is 100 to 300x the published market rate; a card proving beside its miner is competitive near 100 GH/s (the economy analysis of 6 October 2026, sections 3.1 and 4.1, approximate beyond the one card measured; ledger E20). Verification is folded into the chain's own proof; you ship no verifier. The job's base fee rises with the backlog, published at the phase 4 job market.
    3. Users who were not paid to arrive. Every miner is a funded wallet. Pools, payout contracts, hardware finance and hashrate forwards have customers before any consumer app does. Block rewards can pay straight to a contract.
    4. A share of fees, with the number stated. 20% of every priority fee goes to the contracts whose code ran, per call frame, to the payee registered at deployment. Libraries are paid at their code address. Factories pass their registration to what they deploy. At launch fee levels this is a property, not an income: a million 100,000-gas calls a day at a 1 gwei tip pays about 7,300 IGN a year, with 1 gwei taken as a billionth of an IGN (the base unit is Open). It grows with traffic and nothing else.
    @@ -578,7 +593,7 @@ body.all .pager{display:none} ShareGoes toWhy 80%The miner who wins the blockPays the hashrate that secures the chain - 20%The proving pool: shard provers and aggregatorsFor a standing prover population that does not have to hash. On the devnet today the coinbase's 20% output goes to an unspendable script tagged igneum-proving-pool-v0 and is burned there. Provers are paid from a separate escrow in the execution state, credited by rule with the same 20% of each blue block's subsidy and released per shard against valid proof records (Implemented, proving v0, since 5 October 2026). Caveat: consensus does not yet verify the carried proof, it checks the record's statement against native execution and its signature, so today a block producer could claim shard pay with a false proof (ledger P21; the in-consensus verifier is built and switches on when the proven share of blocks reads one). Note: unclaimed pool credit is today stranded in the escrow, no rule returns it; the fix rolls an unproven shard's credit into the next proven segment's pool (0.3.16). Open: the single coinbase payout that replaces the burn, and whether it reclaims the share burned so far + 20%The proving pool: shard provers and aggregatorsFor a standing prover population that does not have to hash. On the devnet today the coinbase's 20% output goes to an unspendable script tagged igneum-proving-pool-v0 and is burned there. Provers are paid from a separate escrow in the execution state, credited by rule with the same 20% of each blue block's subsidy and released per shard against valid proof records (Implemented, proving v0, since 5 October 2026). Caveat: consensus does not yet verify the carried proof, it checks the record's statement against native execution and its signature, so today a block producer could claim shard pay with a false proof (ledger P21; the in-consensus verifier is built; its enforcement, verifier_in_consensus and proof_rule_active_from, is Open and is the prerequisite of the no-rescue network exercise, Deliverable 5). Note: unclaimed pool credit is today stranded in the escrow, no rule returns it; the fix rolls an unproven shard's credit into the next proven segment's pool (designed). Open: the single coinbase payout that replaces the burn, and whether it reclaims the share burned so far 0%Treasury, foundation, team or stakeThere is no coin-holder class in consensus and no tax on emission @@ -599,7 +614,7 @@ body.all .pager{display:none}

    Sources: specification sections 2.5 and 5.1 to 5.4; the engineering log for the devnet receipts and the dev-fee count.

    Security after the subsidy

    -

    The cap stays at 4 billion. There is no tail emission. The schedule is a bet, not a measurement: a halving halves emission income overnight if price and fees do nothing. Kaspa's steeper monthly reduction kept its hashrate while its price rose (approximate). Long term, security is paid for by the proving market and by fees. Outside customers buy proofs as dollars-priced work settled in IGN, and 90% of every job goes to the provers who delivered it, so a prover's income does not depend on emission. The table shows the first year in which the block subsidy on its own pays miners less than the power of about 3,000 consumer cards, at three flat prices. The prices are inputs chosen to span two orders of magnitude. The model runs a 300 W card at 124 MH/s on electricity at USD 0.12 per kWh. One rule sits beside the cap. If external proving revenue is under one fifth of the block subsidy over any 90-day window after year 5, the question of a tail reward goes to the miners' signalling vote. The protocol never changes emission by itself.

    +

    The cap stays at 4 billion. There is no tail emission. The schedule is a bet, not a measurement: a halving halves emission income overnight if price and fees do nothing. Kaspa's steeper monthly reduction kept its hashrate while its price rose (approximate). Long term, security has to be paid by fees and, if it is built and bought, the proving market. The external market is Designed, not built, and is out of the numbers below. As designed, outside customers buy proofs as dollars-priced work settled in IGN, and 90% of every job goes to the provers who delivered it. The table shows the first year in which the block subsidy on its own pays miners less than the power of about 3,000 consumer cards, at three flat prices. The prices are inputs chosen to span two orders of magnitude. The model (modelled, 3 October 2026) runs a 300 W card at 124 MH/s on electricity at USD 0.12 per kWh. One rule sits beside the cap. If external proving revenue is under one fifth of the block subsidy over any 90-day window after year 5, the question of a tail reward goes to the miners' signalling vote. The protocol never changes emission by itself.

    @@ -633,26 +648,26 @@ body.all .pager{display:none} - +
    Price per IGNFirst year the subsidy alone pays under the power of 3,000 cardsSubsidy to miners that year
    Block rewardEmission, 80% to the winnerYes
    In-chain provingThe 20% proving pool plus the proving share of every block's gas (on the devnet, paid from the execution-state escrow; see Economics)Yes, mostly
    External proving jobsRollups and apps on other chains, priced in their moneyNo, but the market is small today and is upside, not a promise
    External proving jobsRollups and apps on other chains, priced in their money (Designed, not built)No, but the market is not built and is upside, not a promise

    A block pays its miner whether or not anyone buys a proof that day. The lottery pays 80 percent of every block from emission; proving is the second income, never the only one. Every useful-work chain on record dropped its miners the day the work stopped paying; Igneum’s miners are paid for the block first.

    -

    The size of that third stream today, in numbers: all of Ethereum L1's proving is about USD 36 a day at the September 2026 tracker cost (USD 0.005 a block, 7,200 blocks a day; the tracker figure is a secondary source), against about USD 13,700 a day of Igneum's year-1 emission at USD 0.005 per IGN (31.688 IGN a block, 86,400 blocks a day; the price is an input, not a forecast). So external proving is a small second income at launch and the lottery pays the bills; for proving to become the main income the paid demand would have to grow about 1,000x in dollars (the Horizon lane analysis, 6 October 2026, section 3.11; ledger E19).

    -

    The honest bear-market case rests on cost. A miner's card is already running and the power is often domestic, so Igneum miners' electricity cost in the proving market is close to power. The price they must charge is another matter: the price a prover must charge is the subsidy it forgoes while it proves, which falls as one over network hash, so the edge over data-centre provers appears only once the network's hash is large (near 100 GH/s for a card proving beside its miner) and is nothing more. Which of the two in-chain streams pays more per GPU-second depends on the size of the fleet: on the devnet of 4 October 2026, three machines at 275 million hashes a second, a second of hashing paid about 4.9x a second of proving the pool share; at 10,000 cards the same arithmetic favours proving by about 930x. That is arithmetic on measured devnet rates, approximate, not a market measurement.

    +

    The size of that third stream today, in numbers: all of Ethereum L1's proving is about USD 36 a day at the September 2026 tracker cost (USD 0.005 a block, 7,200 blocks a day; the tracker figure is a secondary source), against about USD 13,700 a day of Igneum's year-1 emission at USD 0.005 per IGN (31.688 IGN a block, 86,400 blocks a day; the price is an input, not a forecast). So external proving is a small second income at launch and the lottery pays the bills; for proving to become the main income the paid demand would have to grow about 1,000x in dollars (the chip and economy analysis of 6 October 2026, section 3.11; ledger E19).

    +

    The honest bear-market case rests on cost. A miner's card is already running and the power is often domestic, so Igneum miners' electricity cost in the proving market is close to power. The price they must charge is another matter: the price a prover must charge is the subsidy it forgoes while it proves, which falls as one over network hash, so the edge over data-centre provers appears only once the network's hash is large (near 100 GH/s for a card proving beside its miner) and is nothing more. Which of the two in-chain streams pays more per GPU-second depends on the size of the fleet: measured on 4 October 2026, three machines at 275 million hashes a second, a second of hashing paid about 4.9x a second of proving the pool share; at 10,000 cards the same arithmetic favours proving by about 930x. That is arithmetic on measured devnet rates, approximate, not a market measurement.

    Hardware

    -

    The dataset starts at 2 GB and grows (the proposed schedule, fixed at the testnet genesis: 2 GB, doubling at years 4, 12 and 28, the average of half a gigabyte a year), so a 4 GB card mines for about four years and an 8 GB card for about twelve, approximate. Every NVIDIA card from 8 GB proves; 12 GB and up mine and prove; 24 GB on the stock server (eleven rented cards, RTX 3060 to RTX 5090, 6 October 2026). NVIDIA and AMD both work, because the mining program is generated for the architecture both share and the proof system is hash-based. Apple's chips are GPUs with unified memory, so Macs mine too, at about a fifth of a flagship card: Measured, 26.7 against 123 million hashes a second, an Apple M5 Max beside an RTX 5090 on the live devnet, 4 October 2026. A Mac is a poor miner per dollar. There is no CPU mining lane, on purpose, because CPU mining is what botnets farm. Nodes, wallets and exchanges need no GPU at all.

    +

    The dataset starts at 2 GB and grows (the proposed schedule, fixed at genesis: 2 GB, doubling at years 4, 12 and 28, the average of half a gigabyte a year), so a 4 GB card mines for about four years and an 8 GB card for about twelve, approximate. NVIDIA cards prove: from 8 GB on the patched server, 12 GB and up beside the miner, 24 GB on the stock server (measured on eleven rented cards, RTX 3060 to RTX 5090, 6 October 2026). NVIDIA and AMD cards both mine, because the mining program is generated for the architecture both share; only NVIDIA cards prove today. Apple's chips are GPUs with unified memory, so Macs mine too, at about a fifth of a flagship card: Measured, 26.7 against 123 million hashes a second, an Apple M5 Max beside an RTX 5090, mining side by side, 4 October 2026. A Mac is a poor miner per dollar. There is no CPU mining lane, on purpose, because CPU mining is what botnets farm. Nodes, wallets and exchanges need no GPU at all.

    What a miner's hour looks like

    -

    The card hashes the lottery continuously. When the client sees a shard or an external job it can win, it switches the card to proving for a few seconds, posts the proof, and goes back to hashing. The client does the switching and the miner sees one balance.

    +

    The card hashes the lottery continuously. On an NVIDIA card, when the client sees a shard it can win (or, once the job market is built, an external job), it switches the card to proving for a few seconds, posts the proof, and goes back to hashing. The client does the switching and the miner sees one balance.

    The protocol carries no fee: no dev fund, no cut to any team. Ember, the miner software, takes an optional 1% dev fee, the way other GPU miners do. One block template in 100 is requested with the dev address instead of yours, by a counter, not a random draw, so it is exactly 1 in 100 and anyone can check it from the source or from the chain. One flag turns it off (--dev-fee 0, a switch in the app, a line in the HiveOS config). The miner prints the fee and the address when it starts. Any other client is welcome.

    One click, for everyone else

    -

    Farm operators get a HiveOS package. Everyone else gets Igneum Ember: install it on Windows, macOS or Linux, press Start, and the card is mining to a key the app made for you. It is the same client with a face on it. Implemented, Ember 0.3.22 (7 October 2026): the app shows the key once and has you save it before mining starts, or takes an address you already have; the dashboard shows each card's hash rate, blocks found and accepted by the node, the node's height and peers, the next hourly program, the finality votes sent, and a proving tile with shards assigned, submitted and paid and the verifier state; the chain label reads Devnet 3 and the welcome screen says nothing is bought or sold; NVIDIA cards are capped at 80% of their default power limit for stability, with a sweep that looks for the best hash per watt, not yet measured on a card; proving the shards the chain assigns is a switch in Settings (proving v0), beside the 1% dev fee switch, finality voting, and signed updates that install themselves at a quiet moment with a switch to turn that off. It shows no earnings in IGN or in any currency, and it has no hardware-wallet path. Roadmap, Designed and not in the app: earnings per block in IGN with the network named, a figure in your currency, mining paused while you game, and a hardware wallet for your key. Ember is downloaded only from this domain, with the version and size on the button and the hash in the signed manifest the app checks. The next section says what is shipped and what is still a design. Nobody from Igneum will ever ask for your seed. Mining never runs in a browser, because browser compute is slow and browser mining has meant malware since Coinhive. The browser is for the dashboard, and for verifying the chain.

    -

    Testnet terms

    -

    No value. Testnet IGN cannot be sold, bought or redeemed, now or at mainnet. There is no airdrop, no points scheme and no promise tied to testnet balances. Mainnet starts from an empty genesis.

    -

    Resets. The chain restarts from a fresh genesis when a consensus rule changes. Every reset is announced at least seven days ahead on the site and in the app. Balances, contracts and history do not carry over. The devnet that runs today resets without notice.

    +

    Farm operators get a HiveOS package. Everyone else gets Igneum Ember: install it on Windows, macOS or Linux, press Start, and the card is mining to a key the app made for you. It is the same client with a face on it. Implemented in Ember (7 October 2026): the app shows the key once and has you save it before mining starts, or takes an address you already have; the dashboard shows each card's hash rate, blocks found and accepted by the node, the node's height and peers, the next hourly program, the finality votes sent, and a proving tile with shards assigned, submitted and paid and the verifier state; the chain label names the devnet and the welcome screen says nothing is bought or sold; NVIDIA cards are capped at 80% of their default power limit for stability, with a sweep that looks for the best hash per watt, not yet measured on a card; proving the shards the chain assigns is a switch in Settings (proving v0), beside the 1% dev fee switch, finality voting, and signed updates that install themselves at a quiet moment with a switch to turn that off. It shows no earnings in IGN or in any currency, and it has no hardware-wallet path. Roadmap, Designed and not in the app: earnings per block in IGN with the network named, a figure in your currency, mining paused while you game, and a hardware wallet for your key. Ember is downloaded only from this domain, with the version and size on the button and the hash in the signed manifest the app checks. The next section says what is shipped and what is still a design. Nobody from Igneum will ever ask for your seed. Mining never runs in a browser, because browser compute is slow and browser mining has meant malware since Coinhive. The browser is for the dashboard, and for verifying the chain.

    +

    Devnet terms

    +

    No value. Devnet IGN cannot be sold, bought or redeemed, now or at mainnet. There is no airdrop, no points scheme and no promise tied to devnet balances. Mainnet starts from an empty genesis.

    +

    Resets. The devnet may restart from a fresh genesis, without notice. Balances, contracts and history do not carry over.

    What the app sends home. The app version, a random machine id made at install, your operating system, the node version, the hash rate, and the app, node and miner logs (which name the address the card mines to). They go to the project's log intake, a service Igneum runs on its host, and are read by the maintainers to find faults. Never your seed phrase, never a key, never a file you did not make with the app. Nothing is sold or shared.

    Wallet set-up for MetaMask: chain id, RPC and the one-click button. The miner software takes an optional 1% fee, off with one flag; the protocol carries no fee to anyone.

    Fair launch, announced

    -

    Launch date and miner software published a month ahead. Pools live on testnet. HiveOS support on day one. The founders mine from genesis like everyone else, with disclosed addresses and the same software. Nobody has coins before block one. The first 30 days of mainnet run on proof of work alone, with no locked checkpoint, while vote weights build; anyone crediting deposits in that month should treat Igneum as plain proof of work with a 12-hour depth.

    +

    Launch date and miner software published a month ahead. Pools live before launch. HiveOS support on day one. The founders mine from genesis like everyone else, with disclosed addresses and the same software. Nobody has coins before block one. The first 30 days of mainnet run on proof of work alone, with no locked checkpoint, while vote weights build; anyone crediting deposits in that month should treat Igneum as plain proof of work with a 12-hour depth.

@@ -682,12 +697,12 @@ body.all .pager{display:none} 1. Compiler race every hourAt every hourly program the worker compiles up to 17 variants of the kernel (unroll, load path, register budget, threads per block), checks each bit for bit against the base kernel, times each for 2 s with mining paused, and keeps the fastest for the hour. The hash output is bit for bit the sameShipped in the Metal and CUDA workers+17.3% on the genesis seed and +21.2% on the hourly seed, Apple M5 Max, Metal, 14 variants, under load, ratios only. The RTX 5090 race is built and not yet run 2. Per-card auto-tune from the fleetEvery race writes a record to the fleet log. The best variant per card model is aggregated and sent back to every machine inside the signed update manifest, so a new card starts from the fleet's best and keeps racingShipped. The fleet is small, so no table yetNo fleet table yet 3. Hash per wattSteps an NVIDIA card's power cap from 100% to 50% of its default in 10% steps, holds each for 60 s, and keeps the best MH per watt. The tile shows live MH per watt. The sweep never restarts the worker, so the hour's program is never lostIn the app for NVIDIA cards. AMD and Apple: not supportedThe first sweep on a card is pending. The RTX 5090 drew 290 W at p95 under a 460 W cap, so the cap did not bind - 4. Template latencyThe miner subscribes to new templates instead of polling, the node builds the next template ahead, and the workers switch without draining the batch. Target under 50 ms from a new block to the card working on it, solo against the local nodeIn 0.3.6Switched p50 46 to 52 ms, p90 118 to 130 ms, 3-node fast-time network, CPU miners, 0 rejected - 5. Never lose a secondThe next hour's program is compiled ahead and swapped in place. The watchdog, the restarts, the CPU re-check of every found hash and per-worker health on every tile keep the card hashingShippedSwap 0.01 ms on the Mac and 0.00 ms on the RTX 5090, 0 rejected, live devnet. Fake-worker guards: trip 0.0 s, worker back in 2.0 s; a silent worker restarted at 60 s + 4. Template latencyThe miner subscribes to new templates instead of polling, the node builds the next template ahead, and the workers switch without draining the batch. Target under 50 ms from a new block to the card working on it, solo against the local nodeShippedSwitched p50 46 to 52 ms, p90 118 to 130 ms, 3-node fast-time network, CPU miners, 0 rejected + 5. Never lose a secondThe next hour's program is compiled ahead and swapped in place. The watchdog, the restarts, the CPU re-check of every found hash and per-worker health on every tile keep the card hashingShippedSwap 0.01 ms on the Mac and 0.00 ms on the RTX 5090, 0 rejected, mining live. Fake-worker guards: trip 0.0 s, worker back in 2.0 s; a silent worker restarted at 60 s 6. Prove it in publicEvery measured rate, with the card, the miner version, the date and the log entry it came from, on one pageLiveThe bench table -

Measured: engineering log, "miner performance: variant racing" (lever 1), "first hourly program swap on the live devnet" and "miner fault guards and the app watchdog" (lever 5), 4 October 2026; the 0.3.6 release plan, the miner-latency gate (lever 4), 5 October 2026; the efficiency-sweep plan, the RTX 5090 log of 4 October 2026 (lever 3). Levers 2 and 3 are shipped code with no fleet measurement yet.

+

Measured: engineering log, "miner performance: variant racing" (lever 1), "first hourly program swap" and "miner fault guards and the app watchdog" (lever 5), 4 October 2026; the miner-latency gate (lever 4), 5 October 2026; the efficiency-sweep plan, the RTX 5090 log of 4 October 2026 (lever 3). Levers 2 and 3 are shipped code with no fleet measurement yet.

The software's fee, not the protocol's

The protocol is fee-free: no dev fund, no fee to any team, foundation or fund. Ember takes a 1% software dev fee, the norm for GPU miners: default-on, switchable, 1 percent of the producer share (the 80% of emission that pays the block's miner; the proving pool is paid per record and carries none of it). One block template in 100 is requested with the dev address instead of yours, by a counter, never a random draw, so it is exactly 1 in 100 and anyone can check it from the source or from the chain. A fee block still carries your vote key, so it still adds to your finality weight. Ember prints the fee and the address when it starts, shows it in Settings next to a switch, and --dev-fee 0 turns it off, as does DEV_FEE=0 in a HiveOS flight sheet. Any other client is welcome.

Measured: engineering log, "the software dev fee measured on a test network", 4 October 2026: 9 fee blocks in 785 from two fee-paying miners, 0 from the control at --dev-fee 0, the chain and the miners' counters equal.

@@ -718,17 +733,17 @@ body.all .pager{display:none} NextTouch ID and Windows Hello to unlock. In progress, not live. Windows build: next -

Source: Igneum Wallet 0.1.1, 5 October 2026, now 0.1.5 on the downloads host (the wallet source: the vault, HD key, finality, QR and updater modules and the README). Verified: the over-the-air path end to end on one Mac against a test manifest. Not yet run: the Windows path, the rollback paths, a Developer ID signature.

+

Source: Igneum Wallet, first shipped 5 October 2026, on the downloads host (the wallet source: the vault, HD key, finality, QR and updater modules and the README). Verified: the over-the-air path end to end on one Mac against a test manifest. Not yet run: the Windows path, the rollback paths, a Developer ID signature.

Governance

-

Igneum is governed by the hashrate that powers it. Pools carry their hashers' votes, so pool concentration is the governance risk, and it is public: on the devnet the top three vote keys held 34.5% of 8,090 blocks on 4 October 2026, measured.

+

Igneum is governed by the hashrate that powers it. Today, as built, pools carry their hashers' votes, so pool concentration is the governance risk, and it is public: on the devnet the top three vote keys held 34.5% of 8,090 blocks on 4 October 2026, measured.

@@ -741,35 +756,21 @@ body.all .pager{display:none} 1. SpecificationUnder way; closes when the specification is out for external reviewMining generator, shard proving, finality rules, written for external review 2. Prove the provingUnder way; closes at its gateMining program prototype on GPU and CPU, shard proving benchmark on consumer cards. So far: an RTX 5090 proves a shard in 10.9 s compressed; a CPU verifies a warp in 0.61 ms (class v2) to 2.1 ms (class v3). A mid-range card has not been measuredA mid-range GPU proves a shard in under 20 s and a CPU verifies a hash in 10 ms - 3. DevnetDevnet 3 live since 7 October 2026; closes at its gateBlockDAG node with the new mining program and EVM execution. Live now on Devnet 3, every upgrade on from block zero: class v4, the era VDF, difficulty v2, finality v3 (first lock 20:02 UK, 7 October 2026), proving v0 and v1, the ladder at rung 0, calibrated fees, Ember 0.3.22 on every machine. Measured on the first devnet on 6 October 2026: proofs landed a median of about 380 s behind the tip (the observer, /live), against the 60 s gate; Devnet 3's proof lag and proven share are read from the observer as the fleet publishes them1 block a second held with proofs under 60 s behind the tip - 4. Finality and job marketCloses when the finality design passes external review and one rollup signs for the testnetSustained-mining finality, external proving jobs, miner client with auto-switchingFinality design passes external review and one rollup signs for testnet - 5. Public testnetArmed: opens on the go wordOne-click miner app on Windows, macOS and Linux, HiveOS, pools, the first rollup as a proving customer, no coin yet1,000 independent miners run 30 days and rollup proofs are delivered on time - 6. Mainnet fair launchAfter the testnet has passed its gate: 1,000 independent miners for 30 days and rollup proofs on timeGenesis with no premine, 30-day ramp. No listing is arranged, promised or sought by the project + 3. DevnetUnder way; closes at its gateBlockDAG node with the new mining program and EVM execution, every upgrade on from block zero: class v4, the era VDF, difficulty v2, finality v3, proving v0 and v1, the ladder at rung 0, calibrated fees, Ember on every machine. The devnet's proof lag and proven share are read from the observer as the fleet publishes them1 block a second held with proofs under 60 s behind the tip + 4. Finality and job marketCloses when the finality design passes external review and one external customer pays for repeat proving jobsSustained-mining finality, external proving jobs, miner client with auto-switchingFinality design passes external review and one external customer pays for repeat jobs of its exact workload + 5. No-rescue network exerciseOpen; starts once proof verification is enforced in consensusNo founder-operated mining, proving, aggregation or distribution. Epoch boundaries crossed, signing interrupted, the network partitioned, major operators removed, hostile proof submissions, independently written clients, a withholding prover replaced. One-click miner, HiveOS, pools, no coin yetSpecified behaviour with no emergency algorithm change and no privileged intervention; 1,000 independent miners run 30 days + 6. Mainnet fair launchAfter phase 5 has passed its gateGenesis with no premine, 30-day ramp. No listing is arranged, promised or sought by the project

Dates slip. Gates do not. Phase two decides everything. If consumer GPUs cannot prove shards fast enough, Igneum says so and does not launch on promises.

-

The 0.3.6 release plan, 5 October 2026

-

The proving activation of 5 October 2026 set the next release. The table is the plan as written; Ember has since reached 0.3.22 (7 October 2026), and each item's state is in the engineering log.

-
- - - - - - - - - -
ItemWhy
Ember's node runs the proof verifierOn 5 October no node on the network ran one, so a producer stored proofs and never paid them. The app now points its node at the shipped prover host
The Windows package ships the prover hostA PC needed a 20-minute setup by hand before it could prove
The proving tile shows the verifier stateA node that relays proofs but never pays them says so, on the tile and on the live page
Template latencyLever 4 above: subscribe, pre-warm, switch without draining. Switched p50 46 to 52 ms on the gate run
Instant jobsA job published to the fleet reached a machine up to 10 minutes later. The app now holds an outbound connection and fetches the moment a job is published, 3 to 6 s expected, not yet measured
Testnet parameters behind a height switchThe testnet identity and the fee floors adopted on 5 October 2026, so the devnet is never forked by a node update
-

Source: the 0.3.6 release plan, 5 October 2026. The latency number is the plan's gate run on a 3-node fast-time network with CPU miners; the instant-jobs latency is an estimate until the first measured row.

Questions miners ask

Kaspa was GPU-mined too, and IceRiver shipped a chip within two years.

-

Kaspa never promised chip resistance, and its hash was one fixed function, simple enough to put on silicon. Igneum's program is different every hour, its dataset grows past any fixed memory, and its program space widens every era, with no human involved. The public benchmark with a leaderboard ships with the public testnet, and its source is public with the repository then, so you run it on your own card and post the number. The in-house adversarial pass and the public benchmark are where a chip design that beats a GPU by more than 2x would show. And if a chip ever appears, miners are the ones who signal the response.

+

Kaspa never promised chip resistance, and its hash was one fixed function, simple enough to put on silicon. Igneum does not claim a chip cannot be built. It assumes one exists and scores it: the chip model states the energy advantage, the economic advantage and the response capability separately. The public benchmark with a leaderboard is owed work, and so is the comparison against operating GPU networks: Ravencoin's KAWPOW, Ergo and Firo's reference miner. No result exists yet. The in-house adversarial pass and the public benchmark are where a stronger chip design would show. And if a chip ever appears, miners are the ones who signal the response.

Don't ASICs make a chain safer?

-

Three parts. First, what the chain asks hash to do. Hash picks who makes the next block. It does not protect history. A checkpoint locks when signatures reach two thirds of the weight of the last 30 days of blocks (specification section 3). A locked checkpoint is never reorganised by any amount of hash: fork choice runs among the tips that pass through every certified checkpoint. Rented hash has no weight today. It can mine blocks. It cannot rewrite anything older than a lock. A renter with 60% of the network holds 0.0% of the vote on day one (the finality simulator, table B); one matching the whole honest network reaches a third of the weight on day 20 and never two thirds. The lock is fast: median 1,018 ms behind the checkpoint on the three-node test network (engineering log, the finality harness), and on the live devnet the first lock came two hours after genesis, once the window was full. Reorganisations under the lock are shallow: at 1, 2 and 5 blocks a second the deepest honest reorganisation measured was 2, 3 and 7 blocks against a determination depth of 20 (ledger F7, round 2, the fast-time network with 100-ms links); across five continents blocks reached every node at p50 343 ms and p99 666 ms (the 12-node cloud network, 4 October 2026).

+

Three parts. First, what the chain asks hash to do. Hash picks who makes the next block. It does not protect history. A checkpoint locks when signatures reach two thirds of the weight of the last 30 days of blocks (specification section 3). A locked checkpoint is never reorganised by any amount of hash: fork choice runs among the tips that pass through every certified checkpoint. Rented hash has no weight today. It can mine blocks. It cannot rewrite anything older than a lock. A renter with 60% of the network holds 0.0% of the vote on day one (the finality simulator, table B); one matching the whole honest network reaches a third of the weight on day 20 and never two thirds. The lock is fast: median 1,018 ms behind the checkpoint on the three-node test network (engineering log, the finality harness), and with a two-hour window the first lock came two hours after genesis, once the window was full. Reorganisations under the lock are shallow: at 1, 2 and 5 blocks a second the deepest honest reorganisation measured was 2, 3 and 7 blocks against a determination depth of 20 (ledger F7, round 2, the fast-time network with 100-ms links); across five continents blocks reached every node at p50 343 ms and p99 666 ms (the 12-node cloud network, 4 October 2026).

Second, the cost the ASIC argument skips. Kaspa's hash went to a handful of chip owners within months: the IceRiver KS0 shipped in July 2023, 17 to 20 months after launch; hashrate went from under 100 PH/s to over 700 PH/s in months and the GPU share was negligible by late 2023 (the ASIC history, row 23, approximate for the share). Bitcoin's hash comes from two manufacturers and a few pools (approximate, from memory). The first chip's owner mines in secret with an edge for months: on Monero, 85% of the hashrate vanished at the April 2018 fork, and chips were found at over 85% again four months after the next fork (row 16). A chip does not add security to a chain; it moves the chain's security to whoever owns the chip first.

Third, the honest part. A young GPU chain's hash is cheap to rent, and we publish the number beside the chain's own. The locks are what make that rental unable to buy a double-spend: a deposit under a lock stays, whatever the renter mines on top. Ethereum Classic (January 2019 and August 2020), Bitcoin Gold (May 2018 and January 2020) and Vertcoin (October to December 2018, December 2019) were reorganised with rented hash (the ASIC history rows 6 and 7 for Bitcoin Gold and Vertcoin; the Ethereum Classic dates approximate, from memory); Verge's 2018 reorganisations used a timestamp flaw in its multi-algorithm rule as well as hash (approximate). On those chains the rented hash rewrote history because nothing but hash held it. Here the same rental mines blocks for its hour and leaves the locks where they were. The exception is stated above: in the first 30 days of mainnet no checkpoint locks, the chain is plain proof of work with a 12-hour depth, and a rental can reorganise inside that depth; anyone crediting deposits in that month treats it so.

@@ -786,9 +787,9 @@ body.all .pager{display:none}

Correct, and it is the first thing the external review will be paid to break. The specification is public; reviewers will be named and paid before gate 3, and the public benchmark carries the metrics they test against. Until then every finality claim here is a design claim backed by simulations and by the devnet, and the chain runs on plain GHOSTDAG without the rule, so it can be fixed without stopping the chain.

Who are you?

One founder, pseudonymous, working with AI systems. The design, the hostile reviews, the code, the simulators and this document were produced that way, and the commit history says so. The software is shipped by Igneum Labs LTD, Unit IH-00-01-01-OF-01, Level 01, Innovation One, Dubai International Financial Centre. The design remains the work of one founder working with AI systems, reviewed in public through the ledger. What that does and does not mean: the measurements are measurements, reproducible from the commands in the engineering log; the simulators are code anyone can run; the design claims stay design claims until people with names have tried to break them. Every criticism the project expects is kept in a ledger with its honest answer, and the entries that were right are marked conceded; the ledger is public at /ledger. No cryptographer is hired yet; the plan budgets one for phases 1 and 2, and external reviewers are named and paid before gate 3.

-

The founders mine from genesis with disclosed addresses and the same software as everyone else, and hold no coins before block one. The team is pseudonymous and there is no team page. The mining addresses are published at the public testnet; the code history is published with the repository.

+

The founders mine from genesis with disclosed addresses and the same software as everyone else, and hold no coins before block one. The team is pseudonymous and there is no team page. The mining addresses are published before launch; the code history is published with the repository.

Where is the miner?

-

On the devnet now. Igneum Ember runs on Windows, macOS and Linux, a HiveOS package exists, and the devnet's coins have no value. The public benchmark with a leaderboard ships with the public testnet. Pools come with it. The public testnet is armed: three seed nodes and the public RPC are up, and it opens on the go word. All of it before any coin exists. Nothing is asked of a miner before they can run something. The Ember section says what is shipped and what is still owed.

+

On the devnet now. Igneum Ember runs on Windows, macOS and Linux, a HiveOS package exists, and the devnet's coins have no value. The public benchmark with a leaderboard and the pools are owed before launch. All of it before any coin exists. Nothing is asked of a miner before they can run something. The Ember section says what is shipped and what is still owed.

Will my card still pay in a bear market?

Block reward and in-chain proving move with the price. Proving for other chains is priced in the customer's money, and it is a small market today. What Igneum can promise is that its miners' electricity cost in that market is close to power, because the card is already running on domestic power; the price they must charge is the subsidy they forgo, which falls as one over network hash. That is an edge over data-centre provers at scale and nothing more.

@@ -800,7 +801,7 @@ body.all .pager{display:none}

What does a one-minute proof mean for my app?

Nothing you wait for. Your transaction executes in about a second. A miner lock arrives in about two minutes and that is the finality your contract sees. The proof follows and makes the state unforgeable. Liquidations and trades act on executed state at once, as on any chain. A bridge built on Igneum waits for the lock, about two minutes.

Which stablecoin, and is there liquidity?

-

None is bridged at genesis, and no bridge is official. The project will ask Circle for native USDC during the public testnet; whether it is issued is Circle's decision. Anyone may run a bridge at their own risk until the proof bridge arrives with the consensus proof in phase two. The DEX is seeded at launch by the founders' own mined coins and by miners, and every miner is a funded wallet.

+

None is bridged at genesis, and no bridge is official. The project will ask Circle for native USDC before launch; whether it is issued is Circle's decision. Anyone may run a bridge at their own risk until the proof bridge arrives with the consensus proof in phase two. The DEX is seeded at launch by the founders' own mined coins and by miners, and every miner is a funded wallet.

What do I get for being early?

20% of the priority fee on every transaction that runs your code, paid to you every block, which at launch fee levels is small and stated as such above. A place in the wallet's Apps tab and the explorer from day one. First access to the proving precompile and the job market. And a user base that was not paid to arrive: the miners.

How do I deploy?

@@ -830,7 +831,7 @@ body.all .pager{display:none}

Here are the limits, stated before anyone else states them.

Everything in this document is subject to the gates on the roadmap. Nothing in it is an offer to sell anything. Found an error, or a criticism this document does not answer? Email hello@igneum.network, or open an issue on the public specification repository: git.igneum.network/igneum-network/spec/issues. Post reaches Igneum Labs LTD, Unit IH-00-01-01-OF-01, Level 01, Innovation One, Dubai International Financial Centre.

diff --git a/site/randomx.html b/site/randomx.html index 44714401e..366e99e61 100644 --- a/site/randomx.html +++ b/site/randomx.html @@ -250,15 +250,15 @@ - + - +
Hardware it is built forCPUs. GPUs run it badly on purposeGPUs. Any card, any vendor. Bit-exact on Apple, NVIDIA and AMD, measured
Random programPer hash, interpreted in a virtual machinePer hour, compiled to native GPU code. Per hash, the 128 dataset addresses change with the nonce
DatasetAbout 2 GB, the same size since 2019, approximate2 GB, growing (the proposed schedule, fixed at the testnet genesis: 2 GB, doubling at years 4, 12 and 28); a 4 GB card mines about four years, an 8 GB card about twelve, approximate
DatasetAbout 2 GB, the same size since 2019, approximate2 GB, growing (the proposed schedule, fixed at genesis: 2 GB, doubling at years 4, 12 and 28); a 4 GB card mines about four years, an 8 GB card about twelve, approximate
Light verification256 MB cache on a CPU, milliseconds256 MB cache on a CPU (512 MB from year 4), one warp under 10 ms, the gate. Measured 2.1 ms on one Apple M5 Max core for class v3 (3.4x class v2's 0.61 ms); a 2019-class core not yet
Changes over timeNone. A fixed design, unchanged for seven yearsA new program every hour, its memory pattern with it; era draws and reserved families on a schedule fixed at genesis. Nobody touches it
Seed grindingNot applicable, the program comes from the hash inputClosed by a verifiable delay between seed and program
Useful workNone. Hashing onlyEvery NVIDIA card from 8 GB proves; 12 GB and up mine and prove; 24 GB on the stock server (eleven rented cards, RTX 3060 to RTX 5090, 6 October 2026); they sell proofs to other chains. AMD and Apple cards mine, and a prover for them lands when a zkVM ships one
Useful workNone. Hashing onlyNVIDIA cards prove: from 8 GB on the patched server, 12 GB and up beside the miner, 24 GB on the stock server (measured on eleven rented cards, RTX 3060 to RTX 5090, 6 October 2026). Selling proofs to other chains is Designed, not built. AMD and Apple cards mine and do not prove; a prover for them lands when a zkVM ships one
Track recordAbout seven years with one shipped chip, Bitmain’s Antminer X5 (September 2023), at 1.46x per joule over a desktop CPU; the one announced beyond it, Bitmain’s Antminer X9, was withdrawn in May 2026 with zero units; RandomX v2 released 25 March 2026, activation pendingZero years. Every number above is measured and logged with the commands that produced it. The specification, reference hash, test vectors and simulators are public now (git.igneum.network/igneum-network/spec). The node, the miner and the wallet follow to the same host as the repository is published
-

Measured so far: the same hourly program, generated on an Apple M5 Max, compiled by Apple's Metal and NVIDIA's CUDA on an RTX 5090, produced identical hashes on both, 192 of 192 across two programs. On a 1 GB dataset the 5090 ran at about 228 million hashes a second and the Mac at about 45 million, both bound by random memory access rather than arithmetic. Those are prototype figures, not mining rates. The first prototype dataset was a closed-form function, and a miner could compute items instead of loading them: measured 111x faster that way on the Mac. The 256 MB cache construction replaced it on 3 October 2026. With the cache, computing items on the fly runs 4.8x slower than loading them, measured on the Mac, and the honest rate is unchanged on both vendors. Open: the same shortcut ratio on NVIDIA and on a discrete AMD card, and the time-memory trade-off between the two measured points. Inside the 5090's 96 MB cache the same program ran nearly six times faster, which is why the dataset starts at 2 GB and grows. On 4 October 2026 the live devnet crossed an hourly program change on all three vendors with no pause and no rejected block: a Mac at 26.7 million hashes a second, an RTX 5090 at 123 million and an integrated AMD chip at 2.7 million, every hash doing 128 distinct reads of the memory-hard dataset.

+

Measured so far: the same hourly program, generated on an Apple M5 Max, compiled by Apple's Metal and NVIDIA's CUDA on an RTX 5090, produced identical hashes on both, 192 of 192 across two programs. On a 1 GB dataset the 5090 ran at about 228 million hashes a second and the Mac at about 45 million, both bound by random memory access rather than arithmetic. Those are prototype figures, not mining rates. The first prototype dataset was a closed-form function, and a miner could compute items instead of loading them: measured 111x faster that way on the Mac. The 256 MB cache construction replaced it on 3 October 2026. With the cache, computing items on the fly runs 4.8x slower than loading them, measured on the Mac, and the honest rate is unchanged on both vendors. Open: the same shortcut ratio on NVIDIA and on a discrete AMD card, and the time-memory trade-off between the two measured points. Inside the 5090's 96 MB cache the same program ran nearly six times faster, which is why the dataset starts at 2 GB and grows. On 4 October 2026 a live network crossed an hourly program change on all three vendors with no pause and no rejected block: a Mac at 26.7 million hashes a second, an RTX 5090 at 123 million and an integrated AMD chip at 2.7 million, every hash doing 128 distinct reads of the memory-hard dataset.

How mining worksThe chip modelBuilt on the shoulders
diff --git a/tools/ci/ledger-text-check.mjs b/tools/ci/ledger-text-check.mjs index c5514dbdc..4ae77f7a7 100644 --- a/tools/ci/ledger-text-check.mjs +++ b/tools/ci/ledger-text-check.mjs @@ -17,10 +17,8 @@ const REQUIRED = { ['X35', 'its security does not rely on identifying that hardware or retiring it through emergency changes'], ], 'litepaper.html': [ - ['X3', 'Live rows arrive with the public testnet.'], ['G4', 'admin keys in consensus'], ['X8', 'No listing is arranged, promised or sought by the project'], - ['X31', 'The public testnet is armed: three seed nodes and the public RPC are up, and it opens on the go word.'], ['C2', 'Monero has run on RandomX since November 2019'], ['X34', 'was withdrawn in mid-May 2026 before any unit shipped; RandomX 2.0 shipped on 25 March 2026'], ['X36', 'Bitmain opened Antminer X9 pre-orders on 26 December 2025 for July 2026 delivery, then withdrew the product in mid-May 2026 and refunded buyers before any unit shipped; none has been independently benchmarked.'],