Dev Builds » 20260819-1620

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 06:45:50 582725 4000 1812 310 1878 +137.17 ± 4.48 0 18 513 1418 51 +319.35 ± 15.0
ncm-dbt-02 06:47:05 591112 3990 1788 293 1909 +136.84 ± 4.48 0 25 492 1436 42 +321.43 ± 15.36
ncm-dbt-03 06:47:23 578689 3998 1785 276 1937 +137.96 ± 4.22 0 12 495 1463 29 +330.51 ± 15.26
ncm-dbt-04 06:46:51 566271 4012 1822 319 1871 +136.82 ± 4.47 0 20 511 1427 48 +319.22 ± 15.04
ncm-dbt-05 06:46:55 604067 4000 1818 281 1901 +140.72 ± 4.4 0 18 474 1461 47 +334.1 ± 15.63
20000 9025 1479 9496 +137.9 ± 1.97 0 93 2485 7205 217 +324.84 ± 6.81

Test Detail

ID Host Base NPS Games WLD Standard Elo Ptnml(0-2) Gamepair Elo CLI PGN
483916 ncm-dbt-04 564605 12 6 1 5 +154.02 ± 64.21 0 0 1 5 0 +416.21 ± 496.36
483915 ncm-dbt-02 592029 490 224 32 234 +143.82 ± 12.46 0 1 58 179 7 +342.13 ± 45.16
483914 ncm-dbt-01 583237 500 228 37 235 +139.81 ± 13.07 0 5 55 184 6 +330.23 ± 46.62
483913 ncm-dbt-03 581485 498 220 28 250 +141.25 ± 12.02 0 0 63 180 6 +335.65 ± 43.02
483912 ncm-dbt-05 601001 500 226 39 235 +136.56 ± 12.18 0 2 63 181 4 +324.17 ± 43.32
483911 ncm-dbt-04 566849 500 223 37 240 +135.76 ± 13.01 0 3 65 175 7 +312.48 ± 42.67
483910 ncm-dbt-02 592418 500 219 47 234 +124.6 ± 12.71 0 3 76 167 4 +282.94 ± 39.21
483909 ncm-dbt-01 583782 500 229 37 234 +140.62 ± 13.04 0 2 63 176 9 +324.17 ± 43.32
483908 ncm-dbt-03 580116 500 227 40 233 +136.56 ± 11.97 0 2 62 183 3 +327.18 ± 43.69
483907 ncm-dbt-05 602561 500 221 32 247 +138.18 ± 11.26 0 1 60 188 1 +339.63 ± 44.35
483906 ncm-dbt-04 566257 500 223 40 237 +133.34 ± 13.08 0 6 59 181 4 +312.48 ± 44.93
483905 ncm-dbt-02 592116 500 220 37 243 +133.34 ± 12.49 0 2 68 175 5 +309.64 ± 41.56
483904 ncm-dbt-01 583824 500 226 40 234 +135.76 ± 12.01 0 1 66 179 4 +321.19 ± 42.12
483903 ncm-dbt-03 579042 500 227 31 242 +143.89 ± 12.09 0 3 52 191 4 +352.8 ± 48.05
483902 ncm-dbt-05 598524 500 232 35 233 +144.71 ± 13.08 0 3 56 182 9 +339.63 ± 46.2
483901 ncm-dbt-04 566217 500 232 42 226 +138.99 ± 11.88 0 1 62 183 4 +333.32 ± 43.57
483900 ncm-dbt-01 583531 500 229 39 232 +138.99 ± 13.1 0 1 68 171 10 +315.35 ± 41.44
483899 ncm-dbt-02 589625 500 230 33 237 +144.71 ± 12.68 0 4 51 189 6 +349.43 ± 48.52
483898 ncm-dbt-05 598347 500 226 32 242 +142.25 ± 12.37 0 2 58 184 6 +339.63 ± 45.29
483897 ncm-dbt-03 576987 500 227 27 246 +147.19 ± 11.47 0 2 49 196 3 +370.41 ± 49.56
483896 ncm-dbt-04 565902 500 230 37 233 +141.44 ± 12.2 0 2 58 185 5 +339.63 ± 45.29
483895 ncm-dbt-01 578712 500 225 53 222 +124.6 ± 12.53 0 1 81 163 5 +280.42 ± 37.62
483894 ncm-dbt-02 588472 500 234 40 226 +142.25 ± 11.95 0 3 53 191 3 +349.43 ± 47.56
483893 ncm-dbt-05 598171 500 231 35 234 +143.89 ± 11.87 0 0 60 184 6 +346.11 ± 44.19
483892 ncm-dbt-03 580696 500 226 33 241 +141.44 ± 11.77 0 1 59 186 4 +342.85 ± 44.76
483891 ncm-dbt-04 568593 500 220 41 239 +130.14 ± 12.78 0 3 70 172 5 +298.62 ± 41.01
483890 ncm-dbt-01 581610 500 228 34 238 +142.25 ± 11.95 0 2 56 188 4 +346.12 ± 46.15
483889 ncm-dbt-02 590996 500 208 31 261 +128.55 ± 12.43 0 4 67 177 2 +301.33 ± 42.04
483888 ncm-dbt-05 603052 500 228 37 235 +139.81 ± 12.26 0 2 60 183 5 +333.32 ± 44.47
483887 ncm-dbt-03 575191 500 212 41 247 +123.81 ± 12.16 0 1 80 166 3 +282.94 ± 37.88
483886 ncm-dbt-04 562301 500 223 31 246 +140.62 ± 12.23 0 2 59 184 5 +336.46 ± 44.87
483885 ncm-dbt-01 582987 500 223 41 236 +132.54 ± 12.52 0 2 69 174 5 +306.84 ± 41.23
483884 ncm-dbt-02 590996 500 231 40 229 +139.81 ± 13.07 0 3 61 178 8 +324.17 ± 44.15
483883 ncm-dbt-05 604847 500 228 34 238 +142.25 ± 13.37 0 4 57 180 9 +330.23 ± 45.79
483882 ncm-dbt-04 567007 500 231 48 221 +133.34 ± 13.99 0 2 76 159 13 +288.06 ± 39.11
483881 ncm-dbt-03 575883 500 222 47 231 +126.97 ± 11.88 0 1 75 172 2 +295.94 ± 39.27
483880 ncm-dbt-01 584117 500 224 29 247 +143.07 ± 13.15 0 4 55 183 8 +336.46 ± 46.65
483879 ncm-dbt-02 592245 500 222 33 245 +138.18 ± 13.32 0 5 58 180 7 +321.19 ± 45.36
483878 ncm-dbt-05 626039 500 226 37 237 +138.18 ± 13.12 0 4 60 179 7 +321.19 ± 44.57
483877 ncm-dbt-04 568713 500 234 42 224 +140.62 ± 12.02 0 1 61 183 5 +336.46 ± 43.96
483876 ncm-dbt-03 580116 500 224 29 247 +143.07 ± 11.91 0 2 55 189 4 +349.43 ± 46.59

Commit

Commit ID f21610e51a436b69fabdcc89d8fe25ff0c2d5731
Author tony
Date 2026-08-19 16:20:34 UTC
Update network: faster training with AdamW optimizer The resulting net is less sparse and thus a little slower than before, but the better eval prevails at VVLTC. The reduction in training time aids future net experiments. The training occurred at https://github.com/vondele/nettest/pull/448 ...using Tony's AdamW optimizer PR: https://github.com/official-stockfish/nnue-pytorch/pull/511 The main goal was to reduce resource usage during training. As the currently used ranger21 variant uses lookahead and "stable weight decay", the training with adamw is around 5% faster (no stable weight decay) and uses less memory during training and for checkpoints (no lookahead). The other effects came as somewhat of a surprise. It is unclear what exactly causes the lower sparsity, around 50-52% vs 60+% which does cause a significant slowdown. It is also unclear what exactly causes the apparent better scaling. Failed STC nonreg: https://tests.stockfishchess.org/tests/view/6a6e1580c054e285ae029c2e LLR: -2.94 (-2.94,2.94) <-1.75,0.25> Total: 376800 W: 97403 L: 98082 D: 181315 Ptnml(0-2): 1226, 44985, 96614, 44392, 1183 Nodestime STC nonreg: https://tests.stockfishchess.org/tests/view/6a72310d2cf557d58a2e1ced LLR: 2.96 (-2.94,2.94) <-1.75,0.25> Total: 22812 W: 6130 L: 5891 D: 10791 Ptnml(0-2): 89, 2537, 5920, 2766, 94 LTC: https://tests.stockfishchess.org/tests/view/6a7205ef2cf557d58a2e1cb8 Ptnml(0-2): 135, 31150, 81387, 31468, 145 LLR -0.64 <0.00,2.00> LLR: 7.72 <-1.75,0.25> Passed VLTC: https://tests.stockfishchess.org/tests/view/6a7a7b595ebead9b2e786ee7 LLR: 2.95 (-2.94,2.94) <0.00,2.00> Total: 139242 W: 36650 L: 36248 D: 66344 Ptnml(0-2): 13, 13245, 42711, 13631, 21 Passed VVLTC: https://tests.stockfishchess.org/tests/view/6a72cbb02cf557d58a2e1e40 LLR: 2.95 (-2.94,2.94) <0.50,2.50> Total: 71720 W: 18989 L: 18641 D: 34090 Ptnml(0-2): 6, 6229, 23046, 6569, 10 closes https://github.com/official-stockfish/Stockfish/pull/7043 Bench: 2522345
Copyright 2011–2026 Next Chess Move LLC