Dev Builds » 20261024-1814

Use this dev build

NCM plays each Stockfish dev build 20,000 times against Stockfish 15. This yields an approximate Elo difference and establishes confidence in the strength of the dev builds.

Summary

Host Duration Avg Base NPS Games WLD Standard Elo Ptnml(0-2) Gamepair Elo
ncm-dbt-01 07:31:31 586959 4014 1864 263 1887 +146.72 ± 4.22 0 9 436 1514 48 +357.76 ± 16.28
ncm-dbt-02 06:42:56 591757 4000 1832 268 1900 +143.48 ± 4.27 1 13 446 1501 39 +348.6 ± 16.12
ncm-dbt-03 06:43:08 586391 4000 1832 259 1909 +144.41 ± 4.28 0 15 440 1502 43 +350.27 ± 16.23
ncm-dbt-04 06:42:52 571567 4000 1847 280 1873 +143.79 ± 4.3 0 13 453 1488 46 +346.53 ± 15.98
ncm-dbt-05 06:44:30 606533 3986 1865 259 1862 +148.4 ± 4.29 0 7 431 1497 58 +360.31 ± 16.37
20000 9240 1329 9431 +145.35 ± 1.91 1 57 2206 7502 234 +352.63 ± 7.23

Test Detail

ID Host Base NPS Games WLD Standard Elo Ptnml(0-2) Gamepair Elo CLI PGN
487108 ncm-dbt-01 588260 14 7 1 6 +159.07 ± 56.3 0 0 1 6 0 +445.23 ± 880.35 ↓
487107 ncm-dbt-05 619457 486 224 29 233 +147.7 ± 11.64 0 0 53 185 5 +364.88 ± 47.23 ↓
487106 ncm-dbt-02 594280 500 231 44 225 +136.56 ± 12.18 0 1 66 178 5 +321.19 ± 42.12 ↓
487105 ncm-dbt-03 588302 500 230 36 234 +142.25 ± 12.98 0 4 55 184 7 +336.46 ± 46.65 ↓
487104 ncm-dbt-04 573973 500 227 35 238 +140.62 ± 11.14 0 1 57 191 1 +349.43 ± 45.6 ↓
487103 ncm-dbt-01 586816 500 244 31 225 +158.08 ± 12.63 0 2 44 193 11 +389.56 ± 52.48 ↓
487102 ncm-dbt-05 606698 500 238 34 228 +150.51 ± 11.96 0 2 48 194 6 +374.11 ± 50.1 ↓
487101 ncm-dbt-02 592677 500 238 35 227 +149.68 ± 12.01 0 3 46 196 5 +374.11 ± 51.25 ↓
487100 ncm-dbt-03 584159 500 233 37 230 +143.89 ± 12.09 0 1 58 185 6 +346.12 ± 45.18 ↓
487099 ncm-dbt-01 585928 500 241 30 229 +156.39 ± 12.06 0 1 46 194 9 +389.56 ± 51.18 ↓
487098 ncm-dbt-04 571070 500 218 30 252 +137.37 ± 12.15 0 2 62 182 4 +327.18 ± 43.69 ↓
487097 ncm-dbt-05 607695 500 228 30 242 +145.54 ± 12.22 0 2 54 188 6 +352.8 ± 47.05 ↓
487096 ncm-dbt-02 590181 500 216 33 251 +133.34 ± 11.88 0 1 68 178 3 +315.35 ± 41.44 ↓
487095 ncm-dbt-03 586689 500 229 27 244 +148.85 ± 12.27 0 4 45 196 5 +370.41 ± 51.77 ↓
487094 ncm-dbt-01 588217 500 223 31 246 +140.62 ± 11.14 0 1 57 191 1 +349.43 ± 45.6 ↓
487093 ncm-dbt-04 572155 500 237 38 225 +146.36 ± 13.02 0 3 54 184 9 +346.12 ± 47.1 ↓
487092 ncm-dbt-05 602337 500 240 30 230 +155.54 ± 12.78 0 1 50 187 12 +374.11 ± 48.94 ↓
487091 ncm-dbt-02 592720 500 236 33 231 +149.68 ± 13.27 1 1 52 186 10 +359.68 ± 48.01 ↓
487090 ncm-dbt-03 587537 500 227 34 239 +141.44 ± 11.32 0 1 57 190 2 +349.43 ± 45.6 ↓
487089 ncm-dbt-01 585717 500 238 39 223 +146.36 ± 12.81 0 1 59 180 10 +342.85 ± 44.76 ↓
487088 ncm-dbt-04 569949 500 235 27 238 +153.86 ± 11.76 0 0 50 192 8 +381.69 ± 48.77 ↓
487087 ncm-dbt-05 606290 500 230 34 236 +143.89 ± 13.12 0 1 63 175 11 +330.23 ± 43.2 ↓
487086 ncm-dbt-02 586774 500 229 23 248 +152.18 ± 11.64 0 2 45 198 5 +385.59 ± 51.85 ↓
487085 ncm-dbt-04 571511 500 234 40 226 +142.25 ± 11.73 0 1 58 187 4 +346.12 ± 45.18 ↓
487084 ncm-dbt-03 583824 500 230 30 240 +147.19 ± 12.36 0 2 53 188 7 +356.21 ± 47.52 ↓
487083 ncm-dbt-01 586858 500 230 33 237 +144.71 ± 11.83 0 1 56 188 5 +352.8 ± 46.04 ↓
487082 ncm-dbt-05 603993 500 236 38 226 +145.54 ± 11.79 0 1 55 189 5 +356.21 ± 46.49 ↓
487081 ncm-dbt-02 592590 500 222 40 238 +132.54 ± 12.32 0 3 65 179 3 +312.48 ± 42.67 ↓
487080 ncm-dbt-04 568077 500 231 43 226 +137.37 ± 11.73 0 0 66 180 4 +327.17 ± 41.95 ↓
487079 ncm-dbt-03 587707 500 224 34 242 +138.99 ± 12.09 0 2 60 184 4 +333.32 ± 44.47 ↓
487078 ncm-dbt-01 585970 500 231 31 238 +147.19 ± 11.47 0 0 55 190 5 +363.19 ± 46.32 ↓
487077 ncm-dbt-05 604262 500 233 32 235 +148.01 ± 11.19 0 0 53 193 4 +370.41 ± 47.26 ↓
487076 ncm-dbt-04 571873 500 223 24 253 +146.36 ± 11.96 0 2 52 191 5 +359.68 ± 48.01 ↓
487075 ncm-dbt-03 586646 500 229 30 241 +146.36 ± 12.4 0 0 60 181 9 +346.11 ± 44.19 ↓
487074 ncm-dbt-02 592245 500 228 32 240 +143.89 ± 11.42 0 1 55 191 3 +356.21 ± 46.49 ↓
487073 ncm-dbt-01 586223 500 229 37 234 +140.62 ± 12.02 0 2 58 186 4 +339.63 ± 45.29 ↓
487072 ncm-dbt-05 601535 500 236 32 232 +150.51 ± 12.18 0 0 55 186 9 +363.19 ± 46.32 ↓
487071 ncm-dbt-04 573932 500 242 43 215 +146.36 ± 13.61 0 4 54 181 11 +339.63 ± 47.1 ↓
487070 ncm-dbt-03 586266 500 230 31 239 +146.36 ± 11.29 0 1 52 194 3 +366.78 ± 47.92 ↓
487069 ncm-dbt-02 592590 500 232 28 240 +150.51 ± 11.51 0 1 49 195 5 +377.87 ± 49.47 ↓
487068 ncm-dbt-01 588643 500 221 30 249 +139.81 ± 11.62 0 1 60 186 3 +339.63 ± 44.35 ↓

Commit

Commit ID dbe600b7438662e9ade60b31fa0625259dcd0077
Author xiangqishare
Date 2026-10-24 18:14:41 UTC
Unroll NNUE feature updates Clang 23.1.4 leaves the fixed-size loops in `apply_feature_single()` and `apply_feature_pair()` rolled for `x86-64-avx512` and `x86-64-vnni512`. Since `Tile` is passed and returned by value, this causes repeated 1 KiB stack copies during feature updates, including with PGO. Request full unrolling with `#pragma clang loop unroll(full)` for Clang and `#pragma GCC unroll(Tiling::NumRegs / 2)` for GCC, without an AVX512-specific guard. Clang needs its own spelling because Clang 11 does not recognize `#pragma GCC unroll`. The initial Clang-only patch improved NPS by about 41% in a short local non-PGO test on Linux with a Xeon Platinum 8573C, Clang 23.1.4 (`7a941dfb2877`) and full LTO: three alternating before/after pairs, one pinned thread, 256 MiB hash, and four positions at 1M nodes each. No Fishtest run yet. Compatibility checks, all with `-Werror`: | Compiler | Targets and checks | |---|---| | GCC 9.3 | AVX2 and AVX512 accumulator compilation | | Clang 11.1 | AVX2 full LTO build; SSE4.1, AVX2 and AVX512 accumulator compilation; AVX2 unoptimized debug accumulator | | GCC 13.3 | AVX2 and AVX512 accumulator objects unchanged | | Clang 20.1.2 | AVX2 and AVX512 accumulator objects unchanged | | Clang 23.1.5 | SSE4.1, AVX2, AVX512 and VNNI512 accumulator objects unchanged; AVX512 full LTO binary unchanged | “Unchanged” means byte-for-byte identical to the previous PR revision (`b63a546`). Both full builds retain bench `1559611` and identical scores, PVs and node counts across all 51 default positions at 20k nodes. These compatibility checks are not new speed measurements. Prepared with assistance from OpenAI Codex. closes https://github.com/official-stockfish/Stockfish/pull/7251 No functional change
Copyright 2011–2026 Next Chess Move LLC