Dev Builds » 20210518-1606

You are viewing an old NCM Stockfish dev build test. You may find the most recent dev build tests using Stockfish 15 as the baseline here.

Use this dev build

NCM plays each Stockfish dev build 20,000 times against Stockfish 7. This yields an approximate Elo difference and establishes confidence in the strength of the dev builds.

Summary

Host Duration Avg Base NPS Games Wins Losses Draws Elo
ncm-et-3 08:12:10 1969795 3351 2750 14 587 +398.21 ± 14.06
ncm-et-4 08:12:20 1965393 3360 2706 15 639 +382.56 ± 13.46
ncm-et-9 08:12:27 1956054 3301 2679 15 607 +388.59 ± 13.82
ncm-et-10 08:12:40 1956450 3314 2668 21 625 +380.48 ± 13.63
ncm-et-13 08:12:14 1964713 3362 2739 16 607 +391.5 ± 13.82
ncm-et-15 08:12:07 1957692 3313 2708 14 591 +394.79 ± 14.01
20001 16250 95 3656 +389.27 ± 5.62

Test Detail

ID Host Started (UTC) Duration Base NPS Games Wins Losses Draws Elo CLI PGN
146961 ncm-et-9 2021-05-19 08:11 00:45:13 1958013 301 253 1 47 +421.01 ± 50.71
146960 ncm-et-10 2021-05-19 08:09 00:47:15 1954349 314 254 2 58 +384.17 ± 45.42
146959 ncm-et-15 2021-05-19 08:08 00:47:41 1946320 313 253 1 59 +386.69 ± 44.94
146958 ncm-et-3 2021-05-19 08:04 00:51:41 1974383 351 291 3 57 +402.46 ± 45.89
146957 ncm-et-4 2021-05-19 08:04 00:52:14 1962008 360 292 3 65 +384.39 ± 42.85
146956 ncm-et-13 2021-05-19 08:02 00:53:17 1964152 362 286 0 76 +372.3 ± 39.26
146955 ncm-et-9 2021-05-19 06:56 01:13:54 1962615 500 399 2 99 +375.98 ± 34.46
146954 ncm-et-10 2021-05-19 06:54 01:14:17 1950554 500 411 4 85 +395.65 ± 37.35
146953 ncm-et-15 2021-05-19 06:54 01:13:31 1967590 500 407 2 91 +391.57 ± 36.01
146952 ncm-et-4 2021-05-19 06:50 01:12:53 1972655 500 401 2 97 +379.78 ± 34.83
146951 ncm-et-3 2021-05-19 06:50 01:13:36 1960320 500 403 3 94 +381.7 ± 35.44
146950 ncm-et-13 2021-05-19 06:48 01:13:26 1960928 500 407 0 93 +395.65 ± 35.48
146949 ncm-et-9 2021-05-19 05:41 01:14:26 1948435 500 409 1 90 +397.72 ± 36.17
146948 ncm-et-10 2021-05-19 05:40 01:13:20 1962617 500 411 4 85 +395.65 ± 37.35
146947 ncm-et-15 2021-05-19 05:39 01:14:16 1957263 500 408 3 89 +391.57 ± 36.46
146946 ncm-et-3 2021-05-19 05:37 01:12:17 1969710 500 404 1 95 +387.57 ± 35.16
146945 ncm-et-13 2021-05-19 05:34 01:13:15 1961537 500 407 4 89 +387.57 ± 36.48
146944 ncm-et-4 2021-05-19 05:34 01:15:37 1956027 500 410 1 89 +399.81 ± 36.38
146943 ncm-et-9 2021-05-19 04:26 01:14:24 1962173 500 393 0 107 +368.58 ± 32.96
146942 ncm-et-10 2021-05-19 04:25 01:14:30 1954961 500 399 2 99 +375.98 ± 34.46
146941 ncm-et-15 2021-05-19 04:24 01:13:50 1958016 500 404 7 89 +375.98 ± 36.45
146940 ncm-et-3 2021-05-19 04:23 01:13:13 1972344 500 407 4 89 +387.57 ± 36.48
146939 ncm-et-4 2021-05-19 04:21 01:12:08 1971269 500 396 2 102 +370.41 ± 33.92
146938 ncm-et-13 2021-05-19 04:20 01:13:30 1961538 500 396 5 99 +364.98 ± 34.53
146937 ncm-et-9 2021-05-19 03:11 01:14:18 1953642 500 409 4 87 +391.57 ± 36.91
146936 ncm-et-10 2021-05-19 03:10 01:14:10 1961235 500 408 0 92 +397.72 ± 35.69
146935 ncm-et-15 2021-05-19 03:09 01:14:42 1955422 500 422 0 78 +429.05 ± 38.92
146934 ncm-et-3 2021-05-19 03:09 01:13:27 1973276 500 417 1 82 +415.05 ± 37.98
146933 ncm-et-13 2021-05-19 03:07 01:12:25 1975452 500 418 1 81 +417.32 ± 38.22
146932 ncm-et-4 2021-05-19 03:07 01:13:36 1960071 500 392 1 107 +364.98 ± 33.02
146931 ncm-et-10 2021-05-19 01:55 01:14:06 1955567 500 393 2 105 +364.98 ± 33.41
146930 ncm-et-9 2021-05-19 01:55 01:15:13 1957257 500 411 3 86 +397.72 ± 37.12
146929 ncm-et-3 2021-05-19 01:55 01:13:21 1967586 500 413 1 86 +406.2 ± 37.04
146928 ncm-et-4 2021-05-19 01:54 01:12:26 1971571 500 405 4 91 +383.64 ± 36.06
146927 ncm-et-15 2021-05-19 01:53 01:15:05 1958930 500 394 0 106 +370.41 ± 33.12
146926 ncm-et-13 2021-05-19 01:53 01:13:10 1965541 500 408 4 88 +389.56 ± 36.69
146925 ncm-et-3 2021-05-19 00:40 01:14:35 1970949 500 415 1 84 +410.58 ± 37.5
146924 ncm-et-15 2021-05-19 00:40 01:13:02 1960309 500 420 1 79 +421.93 ± 38.73
146923 ncm-et-10 2021-05-19 00:40 01:15:02 1955873 500 392 7 101 +354.5 ± 34.18
146922 ncm-et-4 2021-05-19 00:40 01:13:26 1964152 500 410 2 88 +397.72 ± 36.65
146921 ncm-et-9 2021-05-19 00:40 01:14:59 1950245 500 405 4 91 +383.64 ± 36.06
146920 ncm-et-13 2021-05-19 00:40 01:13:11 1963847 500 417 2 81 +412.8 ± 38.27

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 2011–2024 Next Chess Move LLC