Dev Builds » 20260402-1832

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 04:17:17 583142 2542 1115 225 1202 +127.01 ± 5.64 0 21 355 879 16 +293.05 ± 18.08
ncm-dbt-02 04:15:38 585416 2528 1089 230 1209 +122.94 ± 5.92 1 23 381 834 25 +275.82 ± 17.43
ncm-dbt-03 04:16:50 584969 2546 1117 234 1195 +125.71 ± 5.53 0 12 384 859 18 +287.74 ± 17.28
ncm-dbt-04 04:16:54 566905 2564 1136 234 1194 +127.68 ± 5.63 0 18 365 878 21 +292.76 ± 17.8
ncm-dbt-05 04:16:05 580900 2528 1084 236 1208 +121.24 ± 5.75 1 22 384 842 15 +275.33 ± 17.36
12708 5541 1159 6008 +124.92 ± 2.55 2 96 1869 4292 95 +284.86 ± 7.85

Test Detail

ID Host Base NPS Games WLD Standard Elo Ptnml(0-2) Gamepair Elo CLI PGN
464117 ncm-dbt-05 580489 28 11 2 15 +115.74 ± 49.12 0 0 5 9 0 +265.03 ± 178.38
464116 ncm-dbt-02 584664 28 12 2 14 +129.74 ± 62.4 0 0 5 8 1 +265.03 ± 178.38
464115 ncm-dbt-01 583614 42 18 4 20 +120.4 ± 47.51 0 1 5 15 0 +279.59 ± 184.04
464114 ncm-dbt-03 586646 46 21 5 20 +126.1 ± 44.49 0 1 5 17 0 +298.39 ± 185.8
464113 ncm-dbt-04 565587 64 29 9 26 +112.31 ± 40.23 0 1 11 19 1 +237.45 ± 109.52
464112 ncm-dbt-05 582819 500 213 45 242 +121.45 ± 13.31 0 6 74 166 4 +273.0 ± 39.95
464111 ncm-dbt-02 586943 500 225 45 230 +130.94 ± 13.51 0 5 67 171 7 +295.94 ± 42.08
464110 ncm-dbt-01 584285 500 220 34 246 +135.76 ± 12.01 0 3 60 185 2 +327.18 ± 44.54
464109 ncm-dbt-03 584411 500 219 47 234 +124.6 ± 12.33 0 1 80 165 4 +282.94 ± 37.88
464108 ncm-dbt-04 565194 500 222 37 241 +134.95 ± 11.83 0 2 63 183 2 +324.17 ± 43.32
464107 ncm-dbt-02 585337 500 214 38 248 +127.76 ± 12.65 0 3 72 171 4 +293.29 ± 40.39
464106 ncm-dbt-05 582652 500 216 44 240 +124.6 ± 11.74 0 2 74 174 0 +293.29 ± 39.69
464105 ncm-dbt-03 584916 500 219 44 237 +126.97 ± 12.66 0 3 73 170 4 +290.66 ± 40.08
464104 ncm-dbt-01 582903 500 217 54 229 +117.55 ± 13.36 0 7 76 164 3 +263.42 ± 39.41
464103 ncm-dbt-04 568832 500 217 50 233 +120.67 ± 12.22 0 3 78 168 1 +277.93 ± 38.66
464102 ncm-dbt-02 586393 500 206 49 245 +112.91 ± 13.22 0 5 87 154 4 +247.41 ± 36.56
464101 ncm-dbt-05 579207 500 207 51 242 +112.14 ± 13.22 0 8 79 162 1 +251.89 ± 38.63
464100 ncm-dbt-03 582193 500 219 50 231 +122.24 ± 12.57 0 3 78 166 3 +277.93 ± 38.66
464099 ncm-dbt-01 584369 500 212 37 251 +126.97 ± 12.85 0 4 71 171 4 +290.66 ± 40.76
464098 ncm-dbt-04 567522 500 227 57 216 +123.02 ± 13.47 0 3 82 157 8 +268.17 ± 37.61
464097 ncm-dbt-02 584285 500 221 50 229 +123.81 ± 13.63 1 3 76 164 6 +277.93 ± 39.29
464096 ncm-dbt-05 579166 500 219 42 239 +128.55 ± 12.63 1 2 68 177 2 +304.07 ± 41.65
464095 ncm-dbt-03 584664 500 220 49 231 +123.81 ± 12.73 0 2 80 163 5 +277.93 ± 38.02
464094 ncm-dbt-04 566060 500 213 43 244 +123.02 ± 13.47 0 8 67 172 3 +280.42 ± 42.05
464093 ncm-dbt-01 582611 500 228 48 224 +130.94 ± 12.76 0 2 72 170 6 +298.62 ± 40.29
464092 ncm-dbt-02 584874 500 211 46 243 +119.11 ± 13.34 0 7 74 166 3 +268.17 ± 39.97
464091 ncm-dbt-05 581070 500 218 52 230 +119.89 ± 13.68 0 4 84 154 8 +258.75 ± 37.2
464090 ncm-dbt-03 586985 500 219 39 242 +130.94 ± 11.96 0 2 68 178 2 +309.64 ± 41.56
464089 ncm-dbt-04 568236 500 228 38 234 +138.99 ± 12.3 0 1 64 179 6 +327.18 ± 42.83
464088 ncm-dbt-01 581070 500 220 48 232 +124.6 ± 12.33 0 4 71 174 1 +290.66 ± 40.76

Commit

Commit ID c41c9bb9733be48b232afd042c9331dde9000e03
Author anematode
Date 2026-04-02 18:32:34 UTC
Add -fno-ipa-cp-clone on GCC Passed non-regression STC at tests.stockfishchess.org/tests/view/69c9b8c3a54395a761335044 (set to gainer bounds, *mea culpa*) There's a major performance penalty when compiling the fish with GCC 16. @Torom 's measurements (similar to mine, but more precise): ``` GCC 15 (master) vs. 16 (master) sf_base = 2549174 +/- 2984 (95%) sf_test = 2435456 +/- 2239 (95%) diff = -113718 +/- 3599 (95%) speedup = -4.46100% +/- 0.141% (95%) GCC 15 (master) vs. 16 (patch) sf_base = 2544107 +/- 3091 (95%) sf_test = 2568837 +/- 3164 (95%) diff = 24729 +/- 4359 (95%) speedup = 0.97204% +/- 0.171% (95%) ``` Based on looking at the disassembly + profiling data of the builds, the root cause appears to be overly aggressive code duplication from constant propagation of various functions especially in `search.cpp`. The worst offender is `Worker::search`, which I think is being cloned and optimized based on the value of `cutNode`. `perf stat` indicates that the instruction cache miss rate goes up by ~50% on the GCC 16 build. closes https://github.com/official-stockfish/Stockfish/pull/6691 No functional change
Copyright 2011–2026 Next Chess Move LLC