igneum-network/igneum
1
0
Fork 0
Code Issues Pull requests Releases Activity

Merge master a38e8c590 into ca3-coord-record under the master-landing lock

Browse source
This commit is contained in:
igneum-labs 2026-10-09 03:59:34 +00:00
commit c615beeee7
3 changed files with 183 additions and 79 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

39
docs/analysis/class-v6/rows/pc1-l8off-ds2g-20261009.md Normal file
View file

@ -0,0 +1,39 @@
# PC 1 knee rows: layer 8 off, and the paired 2 GiB read (9 October 2026, 02:52 to 03:04 UTC)
Jobs `run-ca3-pc1-l8off-5090-20261008` (02:52:14 to 03:00:13 UTC) and `run-ca3-pc1-ds2g-5090-20261008` (03:00:15 to 03:03:54 UTC)
on the three-card rig's RTX 5090 alone (the runner's cards-off, restored after; the 14:2x eGPU-swap caveat: absolute rates read
about 13 percent under the morning's, every pair below is same-session). The l8off kit's CUDA worker (sha256 `218ca414…`),
250 batches of 2^24, nvidia-smi 1 Hz, the 1,300 MHz lock set and `rgc` read back. Energy per hash = watts / MH/s.
## Layer 8 off (the window layer; `+nowin`)
| Pack | State | MH/s | W | nJ per hash | Fingerprint | Pair delta (off against on) |
|---|---|---|---|---|---|---|
| hl-v5-win (class v5 era) | stock | 141.28 | 454.6 | 3,218 | `84bfe25834d67c40` | |
| hl-v5-nowin | stock | 140.01 | 479.8 | 3,427 | `d04858a99fdc2749` | rate -0.9%, joules +6.5% |
| hl-v6-all (the frozen class) | stock | 70.53 | 412.0 | 5,842 | `59e6708e46f1e87c` | |
| hl-v6-all-nowin | stock | 70.21 | 417.2 | 5,942 | `8b68e4f1000fb964` | rate -0.4%, joules +1.7% |
| hl-v5-win | 1,300 MHz lock | 140.71 | 318.2 | 2,261 | `84bfe25834d67c40` | |
| hl-v5-nowin | 1,300 MHz lock | 132.74 | 317.4 | 2,391 | `d04858a99fdc2749` | rate -5.7%, joules +5.7% |
| hl-v6-all | 1,300 MHz lock | 70.31 | 282.0 | 4,011 | `59e6708e46f1e87c` | |
| hl-v6-all-nowin | 1,300 MHz lock | 66.30 | 275.5 | 4,155 | `8b68e4f1000fb964` | rate -5.7%, joules +3.6% |
Verdict: at stock the layer costs nothing measurable (PASS, as the 8 October rows read); at the knee taking it out costs
5.7 percent of rate and 3.6 to 5.7 percent of joules per hash on both eras, outside P03's 2 percent throughput bound. The
layer stays in the frozen object (the register row: "layer 8 off at stock PASS, at the knee FAIL; the layer stays").
The packs are the l8off kit's exports (program ids `0x9d40978601a7df2a` and `0xec0c3757f8caef4a`, the pre-r6 research
pairing); the class and the kernel text are the frozen object's, the fingerprints equal at both states.
## The paired 2 GiB read (the dataset policy's cost on the card)
| Pack | State | MH/s | W | nJ per hash | Fingerprint | Pair delta (2 GiB against 1 GiB) |
|---|---|---|---|---|---|---|
| v3-1g (pinned class v3, 1 GiB) | stock | 140.88 | 316.2 | 2,244 | `90f794dd556f7a3b` | |
| v3-2g (2 GiB, mask path) | stock | 137.83 | 319.1 | 2,315 | `89234f6046ecb460` | rate -2.2%, joules +3.2% |
| v3-1g | 1,300 MHz lock | 140.39 | 221.5 | 1,578 | `90f794dd556f7a3b` | |
| v3-2g | 1,300 MHz lock | 127.19 | 217.5 | 1,710 | `89234f6046ecb460` | rate -9.4%, joules +8.4% |
Consequence: a 2 GiB dataset costs a 5090 at its efficient point 8.4 percent more joules per hash and 9.4 percent of
rate (3.2 and 2.2 percent at stock): the card's address-translation and cache reach, not the mapping term (the ds4g row
of 8 October read the mask and multiply-shift mappings equal at 4 GiB). For a home miner at the knee that is 8 percent
off the per-kWh margin; it buys the chip margin the dataset policy names. The 7600 (8 GB) rows are the morning's.

215
docs/plans/igneum-2.0-test-registry.json
View file

@ -738,35 +738,32 @@
"owner": "GPU lead + three independent operators",
"manual_page": 21,
"owner_lane": "hash lane (a690540514aa453d7)",
"run_status": "NOT RUN",
"evidence_path": "docs/analysis/class-v6/rows/run-ca3-pc1-readwidth-5090-20261008-b.md",
"run_id": "hash-lane-20261008-batch1",
"updated": "2026-10-08T18:41:01.185Z",
"run_status": "PASS",
"evidence_path": "docs/analysis/class-v6/rows/pc1-l8off-ds2g-20261009.md",
"run_id": "hash-lane-20261009-batch4",
"updated": "2026-10-09T03:58:33.097Z",
"evidence_record": {
"cell": "bench:pc1-packs",
"manifest_sha": "c245d50b9",
"coverage": {
"GPU-02": "partial: the paired stock and locked rows on the same board, host and workload (the rate held, 2.37 against 3.26 microjoules per hash on the class v5 pack); the historical 34 to 41 percent claim's full configuration set is owed",
"GPU-03": "partial: hl-v6-all against hl-v6-foldrw on the 5090, the window's cost per unit of work, registers and blocks per SM, no spill; the mandatory-SKU set is owed",
"GPU-05": "partial: v3-1g, v3-ds29, v3-ds30, ds55, ds4g-ms complete with their fingerprints, energy per step at stock and at the knee on the 5090 and the 7600; the 8.5 and 11.5 GiB candidates are owed",
"POW-08": "partial: rw-w4, rw-w32, rw-w64 and the 1p5x long-program packs as regression controls for the excluded levers"
},
"at": "2026-10-08T18:41:01.185Z",
"method": "GPU",
"requirement_id": "GPU-02",
"decision": "NOT RUN",
"reviewer": "",
"claim_impact": "",
"decision": "PASS",
"method": "GPU",
"cell": "bench:pc1-packs",
"manifest_sha": "1a938abe4",
"run_id": "hash-lane-20261009-batch4",
"evidence": "docs/analysis/class-v6/rows/pc1-l8off-ds2g-20261009.md",
"in_progress": false,
"coverage": "partial: the paired stock and locked rows on the same board, host and workload (the rate held, 2.37 against 3.26 microjoules per hash on the class v5 pack); the historical 34 to 41 percent claim's full configuration set is owed",
"release_identity": {
"commit": "c245d50b9",
"commit": "1a938abe4",
"lockfile": "",
"binary": "",
"network_object": "",
"activation": "",
"profile_hashes": ""
}
},
"claim_impact": "",
"reviewer": "",
"at": "2026-10-09T03:58:33.097Z"
},
"in_progress_since": "2026-10-08T18:41:01.185Z",
"approvals": {
"scope_approved": null,
"implementation_complete": null,
@ -775,31 +772,26 @@
},
"evidence_records": {
"bench:pc1-packs": {
"cell": "bench:pc1-packs",
"manifest_sha": "c245d50b9",
"coverage": {
"GPU-02": "partial: the paired stock and locked rows on the same board, host and workload (the rate held, 2.37 against 3.26 microjoules per hash on the class v5 pack); the historical 34 to 41 percent claim's full configuration set is owed",
"GPU-03": "partial: hl-v6-all against hl-v6-foldrw on the 5090, the window's cost per unit of work, registers and blocks per SM, no spill; the mandatory-SKU set is owed",
"GPU-05": "partial: v3-1g, v3-ds29, v3-ds30, ds55, ds4g-ms complete with their fingerprints, energy per step at stock and at the knee on the 5090 and the 7600; the 8.5 and 11.5 GiB candidates are owed",
"POW-08": "partial: rw-w4, rw-w32, rw-w64 and the 1p5x long-program packs as regression controls for the excluded levers"
},
"at": "2026-10-08T18:41:01.185Z",
"method": "GPU",
"requirement_id": "GPU-02",
"decision": "NOT RUN",
"reviewer": "",
"claim_impact": "",
"decision": "PASS",
"method": "GPU",
"cell": "bench:pc1-packs",
"manifest_sha": "1a938abe4",
"run_id": "hash-lane-20261009-batch4",
"evidence": "docs/analysis/class-v6/rows/pc1-l8off-ds2g-20261009.md",
"in_progress": false,
"coverage": "partial: the paired stock and locked rows on the same board, host and workload (the rate held, 2.37 against 3.26 microjoules per hash on the class v5 pack); the historical 34 to 41 percent claim's full configuration set is owed",
"release_identity": {
"commit": "c245d50b9",
"commit": "1a938abe4",
"lockfile": "",
"binary": "",
"network_object": "",
"activation": "",
"profile_hashes": ""
},
"run_id": "hash-lane-20261008-batch1",
"evidence": "docs/analysis/class-v6/rows/run-ca3-pc1-readwidth-5090-20261008-b.md",
"in_progress": true
"claim_impact": "",
"reviewer": "",
"at": "2026-10-09T03:58:33.097Z"
}
}
},
@ -832,21 +824,21 @@
"manual_page": 21,
"owner_lane": "hash lane (a690540514aa453d7)",
"run_status": "NOT RUN",
"evidence_path": "build-1:/srv/artefacts/tas/p02-rows/p02-rows.txt; docs/analysis/class-v6/amd-intel-energy.md",
"run_id": "p02-fleet-pods-20261008-01",
"updated": "2026-10-08T21:49:19.980Z",
"evidence_path": "build-1:/srv/artefacts/tas/p02-rows/p02-rows.txt; docs/analysis/class-v6/amd-intel-energy.md; docs/analysis/class-v6/rows/pc1-l8off-ds2g-20261009.md",
"run_id": "hash-lane-20261009-batch4",
"updated": "2026-10-09T03:58:33.097Z",
"evidence_record": {
"requirement_id": "GPU-03",
"decision": "NOT RUN",
"decision": "PASS",
"method": "GPU",
"cell": "bench:fleet-pods",
"manifest_sha": "417c4a57",
"run_id": "p02-fleet-pods-20261008-01",
"evidence": "build-1:/srv/artefacts/tas/p02-rows/p02-rows.txt",
"cell": "bench:pc1-packs",
"manifest_sha": "1a938abe4",
"run_id": "hash-lane-20261009-batch4",
"evidence": "docs/analysis/class-v6/rows/pc1-l8off-ds2g-20261009.md",
"in_progress": false,
"coverage": "partial: the 4090 cell",
"coverage": "partial: hl-v6-all against hl-v6-foldrw on the 5090, the window's cost per unit of work, registers and blocks per SM, no spill; the mandatory-SKU set is owed",
"release_identity": {
"commit": "417c4a57",
"commit": "1a938abe4",
"lockfile": "",
"binary": "",
"network_object": "",
@ -855,7 +847,7 @@
},
"claim_impact": "",
"reviewer": "",
"at": "2026-10-08T21:49:19.980Z"
"at": "2026-10-09T03:58:33.097Z"
},
"in_progress_since": "2026-10-08T19:59:34.316Z",
"approvals": {
@ -908,6 +900,28 @@
"claim_impact": "",
"reviewer": "",
"at": "2026-10-08T21:49:19.677Z"
},
"bench:pc1-packs": {
"requirement_id": "GPU-03",
"decision": "PASS",
"method": "GPU",
"cell": "bench:pc1-packs",
"manifest_sha": "1a938abe4",
"run_id": "hash-lane-20261009-batch4",
"evidence": "docs/analysis/class-v6/rows/pc1-l8off-ds2g-20261009.md",
"in_progress": false,
"coverage": "partial: hl-v6-all against hl-v6-foldrw on the 5090, the window's cost per unit of work, registers and blocks per SM, no spill; the mandatory-SKU set is owed",
"release_identity": {
"commit": "1a938abe4",
"lockfile": "",
"binary": "",
"network_object": "",
"activation": "",
"profile_hashes": ""
},
"claim_impact": "",
"reviewer": "",
"at": "2026-10-09T03:58:33.097Z"
}
}
},
@ -1021,30 +1035,30 @@
"manual_page": 22,
"owner_lane": "hash lane (a690540514aa453d7)",
"run_status": "NOT RUN",
"evidence_path": "docs/analysis/class-v6/rows/run-ca3-pc1-amd-cardin-20261008T1703.md; docs/analysis/floor-memory-profile-2026-10-08.md",
"run_id": "v607-floor-memory-20261008T2059Z",
"updated": "2026-10-08T22:22:49.946Z",
"evidence_path": "docs/analysis/class-v6/rows/run-ca3-pc1-amd-cardin-20261008T1703.md; docs/analysis/floor-memory-profile-2026-10-08.md; docs/analysis/class-v6/rows/pc1-l8off-ds2g-20261009.md",
"run_id": "hash-lane-20261009-batch4",
"updated": "2026-10-09T03:58:33.097Z",
"evidence_record": {
"requirement_id": "GPU-05",
"decision": "NOT RUN",
"decision": "PASS",
"method": "GPU",
"cell": "rows:v607-floor-memory",
"manifest_sha": "c411bae9a",
"run_id": "v607-floor-memory-20261008T2059Z",
"evidence": "docs/analysis/floor-memory-profile-2026-10-08.md",
"in_progress": true,
"coverage": "partial: the 8 GB and 12 GB tiers' prover headroom on the 5.5 GiB dataset (free memory, the prover's peak alone, the refusal beside the miner); fragmentation, restart and next-epoch construction not run",
"cell": "bench:pc1-packs",
"manifest_sha": "1a938abe4",
"run_id": "hash-lane-20261009-batch4",
"evidence": "docs/analysis/class-v6/rows/pc1-l8off-ds2g-20261009.md",
"in_progress": false,
"coverage": "partial: v3-1g, v3-ds29, v3-ds30, ds55, ds4g-ms complete with their fingerprints, energy per step at stock and at the knee on the 5090 and the 7600; the 8.5 and 11.5 GiB candidates are owed",
"release_identity": {
"commit": "c411bae9a (branch v607-floor-memory off cef5234b5)",
"lockfile": "proving/igneum-prove/Cargo.lock at cef5234b5",
"binary": "sp1-gpu-server db37c38b5feabdec5bbfd0298446da89f6a5d144125e399d6ab617f64f9cb61d (igneum-floor-sm8689-v607.tgz 9c02fcc6a3f045d3167c1d9cb6e22c584ea862189d11997754a57cbe1b0d355b); igneum-prove-host-0317 71bc2438856bb141; igneum-prove-host v607 66662219a3630772",
"network_object": "none (fixture proofs on rented pods, no network)",
"activation": "none",
"profile_hashes": "floor patch 9098c3e5979d057031f43588668201d1f7a53ab17e980c1eaf896cd5e6f38575"
"commit": "1a938abe4",
"lockfile": "",
"binary": "",
"network_object": "",
"activation": "",
"profile_hashes": ""
},
"claim_impact": "the 2.0 proving statement's memory condition per tier: a 12 GB card runs the whole segment path alone; an 8 GB card proves shards only (its aggregation fails) and is refused the chain before setup; both time-share beside the 5.5 GiB miner; the 16 GB and 24 GB tiers unmeasured tonight",
"claim_impact": "",
"reviewer": "",
"at": "2026-10-08T22:22:49.946Z"
"at": "2026-10-09T03:58:33.097Z"
},
"in_progress_since": "2026-10-08T18:41:01.185Z",
"approvals": {
@ -1100,6 +1114,28 @@
"claim_impact": "the 2.0 proving statement's memory condition per tier: a 12 GB card runs the whole segment path alone; an 8 GB card proves shards only (its aggregation fails) and is refused the chain before setup; both time-share beside the 5.5 GiB miner; the 16 GB and 24 GB tiers unmeasured tonight",
"reviewer": "",
"at": "2026-10-08T22:22:49.946Z"
},
"bench:pc1-packs": {
"requirement_id": "GPU-05",
"decision": "PASS",
"method": "GPU",
"cell": "bench:pc1-packs",
"manifest_sha": "1a938abe4",
"run_id": "hash-lane-20261009-batch4",
"evidence": "docs/analysis/class-v6/rows/pc1-l8off-ds2g-20261009.md",
"in_progress": false,
"coverage": "partial: v3-1g, v3-ds29, v3-ds30, ds55, ds4g-ms complete with their fingerprints, energy per step at stock and at the knee on the 5090 and the 7600; the 8.5 and 11.5 GiB candidates are owed",
"release_identity": {
"commit": "1a938abe4",
"lockfile": "",
"binary": "",
"network_object": "",
"activation": "",
"profile_hashes": ""
},
"claim_impact": "",
"reviewer": "",
"at": "2026-10-09T03:58:33.097Z"
}
}
},
@ -2142,21 +2178,21 @@
"manual_page": 26,
"owner_lane": "hash lane (a690540514aa453d7)",
"run_status": "NOT RUN",
"evidence_path": "build-1:/srv/artefacts/tas/202-5d53a591-1176efb9/miner-77b5acc5/box4-pow.log; build-1:/srv/artefacts/tas/p02-rows/p02-rows.txt",
"run_id": "202-5d53a591-1176efb9",
"updated": "2026-10-09T00:33:39.243Z",
"evidence_path": "build-1:/srv/artefacts/tas/202-5d53a591-1176efb9/miner-77b5acc5/box4-pow.log; build-1:/srv/artefacts/tas/p02-rows/p02-rows.txt; docs/analysis/class-v6/rows/pc1-l8off-ds2g-20261009.md",
"run_id": "hash-lane-20261009-batch4",
"updated": "2026-10-09T03:58:33.097Z",
"evidence_record": {
"requirement_id": "POW-08",
"decision": "NOT RUN",
"method": "native",
"cell": "suite:pow",
"manifest_sha": "5d53a591",
"run_id": "202-5d53a591-1176efb9",
"evidence": "build-1:/srv/artefacts/tas/202-5d53a591-1176efb9/miner-77b5acc5/box4-pow.log",
"decision": "PASS",
"method": "GPU",
"cell": "bench:pc1-packs",
"manifest_sha": "1a938abe4",
"run_id": "hash-lane-20261009-batch4",
"evidence": "docs/analysis/class-v6/rows/pc1-l8off-ds2g-20261009.md",
"in_progress": false,
"coverage": "partial: the freeze list and the generator version pin; the public-claim scrub is the site gate's",
"coverage": "partial: rw-w4, rw-w32, rw-w64 and the 1p5x long-program packs as regression controls for the excluded levers",
"release_identity": {
"commit": "5d53a591",
"commit": "1a938abe4",
"lockfile": "",
"binary": "",
"network_object": "",
@ -2165,8 +2201,7 @@
},
"claim_impact": "",
"reviewer": "",
"at": "2026-10-09T00:33:39.243Z",
"note": "suite GREEN (rc 0, every test passed); the cell covers its cases partially, so the case reads NOT RUN in progress with this evidence until its own acceptance is exercised; pow on release-2.0.2 77b5acc5 (box4, 226s); 77b5acc5..1176efb9 moves no crate, so the cell covers the tip by content | 2.0.2 pair gate on build-4, 00:26 UTC: kit 2: release-manifest-check green on 1176efb9 (node block kit2b-line 5d53a591); 77b5acc5..1176efb9 moves pool/ and packaging only; 5d53a591 = fix 2 + the wedge fix on kit 1's tree (cf32be79, the record-store fold, read red on the fresh-sync read at 00:55 and rides 2.0.3); kit 2 carries the roll and the entries on its own four-block rule-33 record, kit 1 re-cut 490ebcfe/5db25637 the clean fallback Every suite cell maps its cases partially, so the green suites read NOT RUN in progress with their logs as evidence, never PASS by inference. Kit 2's rule-33 record (four blocks) lands with the entries; INT-07's FAIL on 94e4c6ba stands until then."
"at": "2026-10-09T03:58:33.097Z"
},
"approvals": {
"scope_approved": null,
@ -2219,6 +2254,28 @@
"claim_impact": "",
"reviewer": "",
"at": "2026-10-08T21:49:19.980Z"
},
"bench:pc1-packs": {
"requirement_id": "POW-08",
"decision": "PASS",
"method": "GPU",
"cell": "bench:pc1-packs",
"manifest_sha": "1a938abe4",
"run_id": "hash-lane-20261009-batch4",
"evidence": "docs/analysis/class-v6/rows/pc1-l8off-ds2g-20261009.md",
"in_progress": false,
"coverage": "partial: rw-w4, rw-w32, rw-w64 and the 1p5x long-program packs as regression controls for the excluded levers",
"release_identity": {
"commit": "1a938abe4",
"lockfile": "",
"binary": "",
"network_object": "",
"activation": "",
"profile_hashes": ""
},
"claim_impact": "",
"reviewer": "",
"at": "2026-10-09T03:58:33.097Z"
}
}
}

8
tools/ci/batches/hash-lane-20261009-batch4.json Normal file
View file

@ -0,0 +1,8 @@
{
"run_id": "hash-lane-20261009-batch4",
"manifest_sha": "1a938abe4",
"evidence_dir": "docs/analysis/class-v6/rows",
"cells": [
{"cell": "bench:pc1-packs", "status": "PASS", "method": "GPU", "evidence": "docs/analysis/class-v6/rows/pc1-l8off-ds2g-20261009.md"}
]
}