Dev Builds » 20260504-0632

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 06:40:49 582883 4006 1789 314 1903 +134.23 ± 4.56 1 21 533 1398 50 +309.6 ± 14.72
ncm-dbt-02 06:41:33 586014 3998 1755 316 1927 +130.92 ± 4.46 0 25 543 1398 33 +303.62 ± 14.59
ncm-dbt-03 06:39:58 585058 4000 1772 357 1871 +128.45 ± 4.58 0 33 556 1374 37 +293.95 ± 14.43
ncm-dbt-04 06:40:04 566685 3996 1774 346 1876 +129.89 ± 4.51 0 31 540 1395 32 +300.45 ± 14.65
ncm-dbt-05 06:39:16 582728 4000 1767 330 1903 +130.64 ± 4.41 0 18 561 1387 34 +302.36 ± 14.32
20000 8857 1663 9480 +130.82 ± 2.01 1 128 2733 6952 186 +301.94 ± 6.5

Test Detail

ID Host Base NPS Games WLD Standard Elo Ptnml(0-2) Gamepair Elo CLI PGN
467566 ncm-dbt-01 582778 6 3 1 2 +120.22 ± 109.7 0 0 1 2 0 +279.24 ± 576.89
467565 ncm-dbt-04 565351 496 218 46 232 +125.69 ± 13.5 0 6 69 168 5 +283.83 ± 41.44
467564 ncm-dbt-02 585042 498 217 41 240 +128.32 ± 12.86 0 5 66 175 3 +297.8 ± 42.41
467563 ncm-dbt-03 585211 500 217 45 238 +124.6 ± 12.53 0 4 72 172 2 +288.06 ± 40.46
467562 ncm-dbt-05 583279 500 212 42 246 +123.02 ± 12.56 0 2 80 164 4 +277.93 ± 38.02
467561 ncm-dbt-01 581860 500 229 35 236 +142.25 ± 13.18 0 2 62 176 10 +327.18 ± 43.69
467560 ncm-dbt-04 564880 500 225 46 229 +130.14 ± 12.97 0 4 68 173 5 +298.62 ± 41.71
467559 ncm-dbt-02 586477 500 211 45 244 +119.89 ± 12.79 0 4 79 164 3 +270.57 ± 38.47
467558 ncm-dbt-03 585295 500 222 51 227 +123.81 ± 12.92 0 5 72 170 3 +282.94 ± 40.5
467557 ncm-dbt-01 583950 500 231 36 233 +143.07 ± 11.69 0 1 57 188 4 +349.43 ± 45.6
467556 ncm-dbt-05 581777 500 227 45 228 +132.54 ± 13.1 0 4 66 174 6 +304.07 ± 42.38
467555 ncm-dbt-02 584285 500 228 36 236 +140.62 ± 12.23 0 1 62 181 6 +333.32 ± 43.57
467554 ncm-dbt-04 568951 500 223 36 241 +136.56 ± 11.76 0 2 61 185 2 +330.23 ± 44.08
467553 ncm-dbt-03 585169 500 220 39 241 +131.74 ± 12.93 0 3 69 172 6 +301.33 ± 41.33
467552 ncm-dbt-01 579537 500 217 40 243 +128.55 ± 13.19 1 3 68 174 4 +298.62 ± 41.71
467551 ncm-dbt-05 580945 500 220 34 246 +135.76 ± 12.21 0 3 61 183 3 +324.17 ± 44.15
467550 ncm-dbt-04 567997 500 218 48 234 +123.02 ± 12.17 0 2 78 168 2 +282.94 ± 38.55
467549 ncm-dbt-02 586097 500 215 34 251 +131.74 ± 12.34 0 4 63 181 2 +312.48 ± 43.44
467548 ncm-dbt-03 584201 500 227 47 226 +130.94 ± 12.56 0 3 68 175 4 +304.07 ± 41.65
467547 ncm-dbt-01 584117 500 223 33 244 +138.99 ± 12.7 0 4 57 184 5 +330.23 ± 45.79
467546 ncm-dbt-05 577028 500 212 41 247 +123.81 ± 13.1 0 4 76 165 5 +277.93 ± 39.29
467545 ncm-dbt-04 565272 500 220 33 247 +136.56 ± 12.39 0 4 58 185 3 +327.18 ± 45.37
467544 ncm-dbt-03 585885 500 219 44 237 +126.97 ± 13.41 0 6 68 171 5 +288.06 ± 41.76
467543 ncm-dbt-02 587325 500 221 45 234 +127.76 ± 13.02 0 5 68 173 4 +293.29 ± 41.75
467542 ncm-dbt-01 583531 500 225 47 228 +129.35 ± 13.36 0 2 77 162 9 +285.49 ± 38.83
467541 ncm-dbt-05 581943 500 229 47 224 +132.54 ± 12.52 0 2 69 174 5 +306.84 ± 41.23
467540 ncm-dbt-04 568077 500 224 46 230 +129.35 ± 12.99 0 3 72 169 6 +293.29 ± 40.39
467539 ncm-dbt-03 580862 500 229 37 234 +140.62 ± 13.04 0 2 63 176 9 +324.17 ± 43.32
467538 ncm-dbt-01 583531 500 212 42 246 +123.02 ± 12.56 0 3 77 167 3 +280.42 ± 38.93
467537 ncm-dbt-02 584453 500 223 35 242 +137.37 ± 12.15 0 1 65 179 5 +324.17 ± 42.47
467536 ncm-dbt-05 580075 500 221 34 245 +136.56 ± 12.39 0 1 67 176 6 +318.25 ± 41.77
467535 ncm-dbt-04 567680 500 216 45 239 +123.81 ± 13.46 0 7 69 170 4 +280.42 ± 41.44
467534 ncm-dbt-03 586435 500 214 53 233 +116.0 ± 13.71 0 8 77 161 4 +256.44 ± 39.15
467533 ncm-dbt-01 581985 500 230 44 226 +135.76 ± 13.01 0 4 62 178 6 +315.35 ± 43.81
467532 ncm-dbt-02 586562 500 224 40 236 +134.15 ± 12.47 0 1 70 173 6 +309.64 ± 40.79
467531 ncm-dbt-05 584622 500 215 36 249 +130.14 ± 11.57 0 0 73 175 2 +306.84 ± 39.69
467530 ncm-dbt-04 565272 500 230 46 224 +134.15 ± 12.67 0 3 65 177 5 +312.48 ± 42.67
467529 ncm-dbt-03 587409 500 224 41 235 +133.34 ± 12.29 0 2 67 177 4 +312.48 ± 41.9
467528 ncm-dbt-01 584664 500 219 36 245 +133.34 ± 13.26 0 2 72 167 9 +298.62 ± 40.29
467527 ncm-dbt-05 592159 500 231 51 218 +130.94 ± 12.17 0 2 69 176 3 +306.84 ± 41.23
467526 ncm-dbt-02 587877 500 216 40 244 +127.76 ± 12.84 0 4 70 172 4 +293.29 ± 41.07

Commit

Commit ID ab8f901d25def6c17f83b7a1b4b84eaec404c0f4
Author anematode
Date 2026-05-04 06:32:42 UTC
Remove small net Failed VVLTC non-regression https://tests.stockfishchess.org/tests/view/69d562b84088e069540a2288 LLR: -2.96 (-2.94,2.94) <-1.75,0.25> Total: 386998 W: 99181 L: 99760 D: 188057 Ptnml(0-2): 35, 35792, 122429, 35203, 40 Failed STC non-regression https://tests.stockfishchess.org/tests/view/69f3c6601e5788938e86a99e LLR: -2.93 (-2.94,2.94) <-1.75,0.25> Total: 33696 W: 8492 L: 8795 D: 16409 Ptnml(0-2): 124, 4209, 8504, 3868, 143 Many thanks to Dubslow, Torom, ces42, Shawn, vondele, Disservin and others for discussion. ## Summary The venerable small net has been around for quite some time now, and while the big net architecture has substantially advanced with TI, the small net has stayed with plain HalfKA. It therefore presents a few burdens: multiple net architectures to maintain, multiple nets to train, and a whole lot of templates to deal with the variable L1 size. Locally I measure a slowdown of -2.5% in NPS with this branch – and it's probably more on non-AVX512 architectures – but a pure slowdown of that magnitude would lead to more dramatic losses (even at VVLTC) than exhibited in the above tests, suggesting that the small net's lower eval quality is deleterious. Bonus: Shawn found this interesting PGN among the VVLTC games: https://lichess.org/study/hvo8jflc/OeTOityv `master` seems to misevaluate the fortress because all positions go to small net (the material difference is larger than the threshold). closes https://github.com/official-stockfish/Stockfish/pull/6796 Bench: 2877007
Copyright 2011–2026 Next Chess Move LLC