Dev Builds » 20210518-1606

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 06:48:14 584034 4030 327 1814 1889 -134.54 ± 4.95 88 1348 542 37 0 -297.47 ± 14.64
ncm-dbt-02 06:47:22 586694 4000 335 1774 1891 -130.84 ± 5.11 92 1303 557 48 0 -283.9 ± 14.44
ncm-dbt-03 06:47:26 586444 4000 323 1839 1838 -138.59 ± 5.17 111 1338 507 44 0 -303.04 ± 15.15
ncm-dbt-04 06:47:44 566817 4000 313 1797 1890 -135.35 ± 5.02 86 1357 513 43 1 -300.99 ± 15.06
ncm-dbt-05 06:48:48 585245 3970 320 1822 1828 -138.32 ± 5.0 94 1349 507 35 0 -307.76 ± 15.14
20000 1618 9046 9336 -135.52 ± 2.26 471 6695 2626 207 1 -298.49 ± 6.65

Test Detail

ID Host Base NPS Games WLD Standard Elo Ptnml(0-2) Gamepair Elo CLI PGN
455712 ncm-dbt-01 583028 30 1 14 15 -161.14 ± 68.32 2 9 4 0 0 -325.08 ± 234.24
455711 ncm-dbt-05 584327 470 33 213 224 -140.2 ± 13.39 6 172 53 4 0 -330.56 ± 47.54
455710 ncm-dbt-02 585211 500 42 222 236 -130.94 ± 13.14 5 175 65 5 0 -301.33 ± 42.75
455709 ncm-dbt-03 588217 500 40 228 232 -137.37 ± 14.08 12 168 66 4 0 -304.07 ± 42.38
455708 ncm-dbt-04 564370 500 34 230 236 -143.89 ± 14.44 14 173 58 5 0 -321.19 ± 45.36
455707 ncm-dbt-01 585084 500 56 234 210 -129.35 ± 14.24 12 158 76 4 0 -277.93 ± 39.29
455706 ncm-dbt-05 606471 500 35 226 239 -139.81 ± 14.55 14 168 63 5 0 -306.84 ± 43.45
455705 ncm-dbt-04 566138 500 45 226 229 -131.74 ± 14.88 11 168 62 9 0 -288.06 ± 43.65
455704 ncm-dbt-02 586308 500 49 223 228 -126.17 ± 14.95 15 149 81 5 0 -261.07 ± 38.02
455703 ncm-dbt-03 586393 500 43 232 225 -138.18 ± 14.41 14 165 67 4 0 -301.33 ± 42.04
455702 ncm-dbt-01 584285 500 35 212 253 -128.55 ± 13.73 7 169 68 6 0 -288.06 ± 41.76
455701 ncm-dbt-05 580157 500 34 230 236 -143.89 ± 13.31 10 179 58 3 0 -333.32 ± 45.35
455700 ncm-dbt-03 585084 500 42 233 225 -139.81 ± 14.01 12 171 63 4 0 -312.48 ± 43.44
455699 ncm-dbt-02 584706 500 50 227 223 -128.55 ± 14.26 12 157 77 4 0 -275.45 ± 39.01
455698 ncm-dbt-04 565902 500 32 215 253 -133.34 ± 14.34 11 167 66 6 0 -293.29 ± 42.41
455697 ncm-dbt-01 583572 500 38 230 232 -140.62 ± 13.43 12 169 68 1 0 -315.35 ± 41.44
455696 ncm-dbt-05 579620 500 44 230 226 -135.76 ± 14.47 14 162 70 4 0 -293.29 ± 41.07
455695 ncm-dbt-03 588899 500 42 228 230 -135.76 ± 12.62 6 176 66 2 0 -315.35 ± 42.24
455694 ncm-dbt-04 568156 500 33 218 249 -134.95 ± 13.03 8 171 69 2 0 -306.84 ± 41.23
455693 ncm-dbt-02 588345 500 44 215 241 -123.81 ± 14.65 11 156 76 7 0 -263.42 ± 39.41
455692 ncm-dbt-01 584076 500 40 219 241 -130.14 ± 14.06 9 167 68 6 0 -288.06 ± 41.76
455691 ncm-dbt-05 584706 500 45 227 228 -132.54 ± 14.87 13 163 67 7 0 -285.49 ± 42.08
455690 ncm-dbt-03 586943 500 39 239 222 -147.19 ± 15.05 17 172 55 6 0 -324.17 ± 46.56
455689 ncm-dbt-02 589368 500 43 220 237 -128.55 ± 14.42 8 170 63 9 0 -285.49 ± 43.31
455688 ncm-dbt-04 567680 500 34 221 245 -136.56 ± 14.1 10 173 61 6 0 -306.84 ± 44.17
455687 ncm-dbt-01 582318 500 45 233 222 -137.37 ± 14.25 14 163 70 3 0 -298.62 ± 41.01
455686 ncm-dbt-03 584117 500 38 227 235 -138.18 ± 15.76 15 170 54 11 0 -298.62 ± 46.39
455685 ncm-dbt-05 581860 500 50 232 218 -132.54 ± 14.7 13 162 69 6 0 -285.49 ± 41.44
455684 ncm-dbt-02 584076 500 49 220 231 -123.81 ± 16.19 19 142 80 9 0 -245.2 ± 38.39
455683 ncm-dbt-04 568275 500 39 230 231 -139.81 ± 13.83 13 167 68 2 0 -309.64 ± 41.56
455682 ncm-dbt-01 584117 500 40 218 242 -129.35 ± 13.54 5 175 63 7 0 -295.94 ± 43.42
455681 ncm-dbt-03 586562 500 37 229 234 -140.62 ± 15.38 18 162 64 6 0 -298.62 ± 43.1
455680 ncm-dbt-05 584958 500 35 231 234 -143.89 ± 13.12 9 181 57 3 0 -336.46 ± 45.77
455679 ncm-dbt-02 587792 500 36 225 239 -138.18 ± 13.69 9 176 60 5 0 -315.35 ± 44.57
455678 ncm-dbt-04 568077 500 46 223 231 -128.55 ± 14.92 10 166 66 7 1 -282.94 ± 42.37
455677 ncm-dbt-01 585126 500 34 226 240 -140.62 ± 14.7 15 167 63 5 0 -306.84 ± 43.45
455676 ncm-dbt-03 585337 500 42 223 235 -131.74 ± 15.53 17 154 72 7 0 -273.0 ± 40.54
455675 ncm-dbt-02 587749 500 22 222 256 -147.19 ± 13.96 13 178 55 4 0 -336.46 ± 46.65
455674 ncm-dbt-04 565941 500 50 234 216 -134.15 ± 13.97 9 172 63 6 0 -301.33 ± 43.45
455673 ncm-dbt-05 579868 500 44 233 223 -138.18 ± 14.41 15 162 70 3 0 -298.62 ± 41.01
455672 ncm-dbt-01 584706 500 38 228 234 -138.99 ± 14.21 12 171 62 5 0 -309.64 ± 43.82

Commit

Commit ID e8d64af1230fdac65bb0da246df3e7abe82e0838
Author Tomasz Sobczyk
Date 2021-05-18 16:06:23 UTC
New NNUE architecture and net Introduces a new NNUE network architecture and associated network parameters, as obtained by a new pytorch trainer. The network is already very strong at short TC, without regression at longer TC, and has potential for further improvements. https://tests.stockfishchess.org/tests/view/60a159c65085663412d0921d TC: 10s+0.1s, 1 thread ELO: 21.74 +-3.4 (95%) LOS: 100.0% Total: 10000 W: 1559 L: 934 D: 7507 Ptnml(0-2): 38, 701, 2972, 1176, 113 https://tests.stockfishchess.org/tests/view/60a187005085663412d0925b TC: 60s+0.6s, 1 thread ELO: 5.85 +-1.7 (95%) LOS: 100.0% Total: 20000 W: 1381 L: 1044 D: 17575 Ptnml(0-2): 27, 885, 7864, 1172, 52 https://tests.stockfishchess.org/tests/view/60a2beede229097940a03806 TC: 20s+0.2s, 8 threads LLR: 2.93 (-2.94,2.94) <0.50,3.50> Total: 34272 W: 1610 L: 1452 D: 31210 Ptnml(0-2): 30, 1285, 14350, 1439, 32 https://tests.stockfishchess.org/tests/view/60a2d687e229097940a03c72 TC: 60s+0.6s, 8 threads LLR: 2.94 (-2.94,2.94) <-2.50,0.50> Total: 45544 W: 1262 L: 1214 D: 43068 Ptnml(0-2): 12, 1129, 20442, 1177, 12 The network has been trained (by vondele) using the https://github.com/glinscott/nnue-pytorch/ trainer (started by glinscott), specifically the branch https://github.com/Sopel97/nnue-pytorch/tree/experiment_56. The data used are in 64 billion positions (193GB total) generated and scored with the current master net d8: https://drive.google.com/file/d/1hOOYSDKgOOp38ZmD0N4DV82TOLHzjUiF/view?usp=sharing d9: https://drive.google.com/file/d/1VlhnHL8f-20AXhGkILujnNXHwy9T-MQw/view?usp=sharing d10: https://drive.google.com/file/d/1ZC5upzBYMmMj1gMYCkt6rCxQG0GnO3Kk/view?usp=sharing fishtest_d9: https://drive.google.com/file/d/1GQHt0oNgKaHazwJFTRbXhlCN3FbUedFq/view?usp=sharing This network also contains a few architectural changes with respect to the current master: Size changed from 256x2-32-32-1 to 512x2-16-32-1 ~15-20% slower ~2x larger adds a special path for 16 valued ClippedReLU fixes affine transform code for 16 inputs/outputs, buy using InputDimensions instead of PaddedInputDimensions this is safe now because the inputs are processed in groups of 4 in the current affine transform code The feature set changed from HalfKP to HalfKAv2 Includes information about the kings like HalfKA Packs king features better, resulting in 8% size reduction compared to HalfKA The board is flipped for the black's perspective, instead of rotated like in the current master PSQT values for each feature the feature transformer now outputs a part that is fowarded directly to the output and allows learning piece values more directly than the previous network architecture. The effect is visible for high imbalance positions, where the current master network outputs evaluations skewed towards zero. 8 PSQT values per feature, chosen based on (popcount(pos.pieces()) - 1) / 4 initialized to classical material values on the start of the training 8 subnetworks (512x2->16->32->1), chosen based on (popcount(pos.pieces()) - 1) / 4 only one subnetwork is evaluated for any position, no or marginal speed loss A diagram of the network is available: https://user-images.githubusercontent.com/8037982/118656988-553a1700-b7eb-11eb-82ef-56a11cbebbf2.png A more complete description: https://github.com/glinscott/nnue-pytorch/blob/master/docs/nnue.md closes https://github.com/official-stockfish/Stockfish/pull/3474 Bench: 3806488
Copyright 2011–2026 Next Chess Move LLC