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

Counter ASIC 3.0 status: job 1 closed (G1 AMD green, the 9070 XT ladder flat to 330,700 ops, watts owed)

Browse source
This commit is contained in:
igneum-josh 2026-10-06 16:58:03 +01:00
parent 01b092d859
commit 4ce9d6589a
1 changed files with 14 additions and 1 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

15
docs/plans/counter-asic-3-status.md
View file

@ -86,6 +86,19 @@ Chip side (approximate: the f = 1 stored-dataset chip of item 1 plus an ALU core
Consequences per tier at N = 100,000: an Apple miner loses 1.5 percent of rate and pays 16 W more (0.56x per watt, per pound unchanged); a 5090 loses 0.2 percent and goes from 350 to 431 W (0.81x per watt), so a 5090 rig pays about 23 percent more electricity for the same blocks; a pool user sees nothing; a small NVIDIA card (4060 class) binds near 100,000 by its ALU budget (model, owed); AMD holds by its budget (owed); every verifier tier is untouched (+0.17 ms per warp).
### Item 8, the RX 9070 XT rows (PC 1 job run-ca3-pc1-amd-g1-shadow-20261006, 15:46:27 to 15:56:54Z, exit 0; the 9070 XT alone through api/cards in every key form (amd:3:gfx1201 is the installed key today), confirmed by the process list, restored in finally and mining again 15 s later; the 5090 and 4070 mining beside it, the 5090 probably clock-locked at about 2,781 MHz from the aborted Ember step; AMD OpenCL 3683.0, device-event time, 2^24 per dispatch)
| N ops per hash | Pack | 9070 XT MH/s (window s) | OpenCL build ms / dataset ms | Watts | Fingerprint = the Mac's |
|---|---|---|---|---|---|
| 930 (class v3) | mx8-devnet-epoch0 | 18.92 (82), 18.92 again at the end | 41 to 51 / 74 | OWED (the ADLX helper ran, 0 samples parsed) | yes |
| 49,700 | sh256x13 | 19.31 (80) | 429 / 74 | OWED | yes |
| 49,700, 64-instruction block | sh64x52 | 19.07 (81) | 342 / 74 | OWED | yes |
| 102,100 (the candidate) | sh256x27 | 19.29 (80) | 428 / 74 | OWED | yes |
| 199,600 | sh256x53 | 19.11 (71) | 424 / 74 | OWED | yes |
| 330,700 | sh256x88 | 19.60 (53) | 418 / 70 | OWED | yes |
Reading: the AMD card never leaves the latency bound on this ladder (its ALU budget is about 650,000 ops per hash); the +2 to 4 percent with the shadow is inside its clock and noise band. Consequence per tier: an AMD RDNA 4 owner loses nothing at the candidate and nothing at 330,700; the N the project can pick stays set by the M5 Max (about 130,000) and the capped 5090 (about 210,000); the one-click AMD miner pays 0.42 s of OpenCL compile per pack at the boundary against 0.05 on class v3, under half a second; the daily build is unchanged at 74 ms. The AMD per-joule row (item 1) and MH per W stay OWED until the sampler reads watts.
### Item 6, the 5090 rows (ca3-reserve b85f0f1; job run-ca3-family-pc2-20261006, 08:42:27 to 08:43:03Z, exit 0, the card quiet, prover off and back on; nvcc 12.8 sm_120; best of 3 runs; every row bit-exact)
| Family | M5 Max, Metal (ratio to the alu chain, 881 G steps/s) | RTX 5090, CUDA (ratio to the alu chain, 7,941 G steps/s) | RX 9070 XT | Native or emulated |
@ -170,7 +183,7 @@ No publish, no manifest, nothing on the live devnet; PC 2 one job at a time with
| Gate | What | State | Evidence |
|---|---|---|---|
| G1 | bit-exact v4 on every vendor against the Mac reference (2^24 fingerprint, self-test) | RUNNING: Metal, Apple OpenCL, CUDA (PC 2); AMD GREEN on the candidate pack (interim) | AMD (PC 1 job run-ca3-pc1-amd-g1-shadow-20261006, started 15:46:27Z, the 9070 XT alone, the 5090 and 4070 mining beside it, the 5090 probably clock-locked at 2,781 MHz from the aborted Ember step): `sh256x27` on gfx1201 (AMD OpenCL 3683.0, build 428 ms) fingerprint 3d2e8245cc084d07 = the Mac's, self-test PASS; 90 dispatches of 2^24 at a mean 870 ms, about 19.3 MH/s against the card's 18.6 to 19.2 on class v3: the AMD tier holds its rate at 100,000 ops per hash. The other packs and the watts rows land with the closing report |
| G1 | bit-exact v4 on every vendor against the Mac reference (2^24 fingerprint, self-test) | RUNNING: Metal, Apple OpenCL, CUDA (PC 2); AMD GREEN (job 1 closed 15:56:54Z exit 0: 7 of 7 packs match: mx8 90f794dd556f7a3b, sh256x13 59ac286fe2a5a9ef, sh256x27 3d2e8245cc084d07, sh256x53 4f824b15cf2b124a, sh256x88 0572522e39a94d8a, sh64x52 9dd010f79d8ca9f4; self-tests PASS) | AMD (PC 1 job run-ca3-pc1-amd-g1-shadow-20261006, started 15:46:27Z, the 9070 XT alone, the 5090 and 4070 mining beside it, the 5090 probably clock-locked at 2,781 MHz from the aborted Ember step): `sh256x27` on gfx1201 (AMD OpenCL 3683.0, build 428 ms) fingerprint 3d2e8245cc084d07 = the Mac's, self-test PASS; 90 dispatches of 2^24 at a mean 870 ms, about 19.3 MH/s against the card's 18.6 to 19.2 on class v3: the AMD tier holds its rate at 100,000 ops per hash. The other packs and the watts rows land with the closing report |
| G2 | the CPU verifier exact on 1,024 random hashes per card | RUNNING (Mac; the 5090 in the same PC 2 job); AMD waits on PC 1 | |
| G3 | the soundness suite green on the class (crate suite; Metal fuzz, edge, stats, determinism) | RUNNING | |
| Verifier benchmark | ms per warp on one core, v2 / x8 / v4 in one session, gate 10 ms | RUNNING | |