Dev Builds » 20210820-0650

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 08:20:37 1958566 3344 2839 5 500 +433.31 ± 15.23
ncm-et-4 08:20:41 1956395 3341 2823 6 512 +428.04 ± 15.05
ncm-et-9 08:20:37 1963478 3336 2818 7 511 +427.4 ± 15.07
ncm-et-10 08:21:11 1949623 3289 2767 7 515 +423.29 ± 15.01
ncm-et-13 08:20:53 1958677 3333 2826 5 502 +431.96 ± 15.2
ncm-et-15 08:21:01 1957945 3357 2863 1 493 +439.65 ± 15.32
20000 16936 31 3033 +430.57 ± 6.17

Test Detail

ID Host Started (UTC) Duration Base NPS Games Wins Losses Draws Elo CLI PGN
150896 ncm-et-10 2021-08-20 20:03 00:44:22 1953895 289 241 0 48 +417.21 ± 49.98
150895 ncm-et-13 2021-08-20 19:57 00:50:36 1954196 333 280 1 52 +421.74 ± 48.1
150894 ncm-et-4 2021-08-20 19:56 00:51:11 1946783 341 295 0 46 +456.27 ± 51.25
150893 ncm-et-9 2021-08-20 19:56 00:51:13 1966318 336 288 0 48 +445.57 ± 50.08
150892 ncm-et-3 2021-08-20 19:54 00:53:00 1955127 344 291 0 53 +431.39 ± 47.52
150891 ncm-et-15 2021-08-20 19:53 00:54:14 1950465 357 304 0 53 +438.36 ± 47.54
150890 ncm-et-10 2021-08-20 18:46 01:16:00 1950396 500 415 1 84 +410.58 ± 37.5
150889 ncm-et-4 2021-08-20 18:41 01:14:21 1960657 500 428 1 71 +441.5 ± 40.95
150888 ncm-et-9 2021-08-20 18:40 01:14:47 1966770 500 409 1 90 +397.72 ± 36.17
150887 ncm-et-3 2021-08-20 18:40 01:13:51 1958014 500 433 0 67 +457.52 ± 42.15
150886 ncm-et-13 2021-08-20 18:39 01:16:34 1953741 500 421 0 79 +426.65 ± 38.66
150885 ncm-et-15 2021-08-20 18:39 01:13:37 1960163 500 437 0 63 +468.95 ± 43.54
150884 ncm-et-10 2021-08-20 17:29 01:17:01 1938773 500 417 2 81 +412.8 ± 38.27
150883 ncm-et-4 2021-08-20 17:27 01:13:42 1962103 500 423 1 76 +429.05 ± 39.52
150882 ncm-et-9 2021-08-20 17:26 01:14:03 1966311 500 429 2 69 +441.5 ± 41.61
150881 ncm-et-3 2021-08-20 17:24 01:15:06 1959403 500 433 0 67 +457.52 ± 42.15
150880 ncm-et-13 2021-08-20 17:24 01:15:11 1961996 500 422 1 77 +426.65 ± 39.25
150879 ncm-et-15 2021-08-20 17:22 01:15:57 1955113 500 432 0 68 +454.76 ± 41.82
150878 ncm-et-10 2021-08-20 16:14 01:14:30 1960008 500 423 1 76 +429.05 ± 39.52
150877 ncm-et-4 2021-08-20 16:11 01:14:29 1964618 500 427 0 73 +441.5 ± 40.29
150876 ncm-et-9 2021-08-20 16:10 01:14:39 1962468 500 430 1 69 +446.7 ± 41.57
150875 ncm-et-13 2021-08-20 16:09 01:13:44 1970495 500 429 1 70 +444.09 ± 41.26
150874 ncm-et-3 2021-08-20 16:08 01:14:52 1960462 500 410 0 90 +401.92 ± 36.1
150873 ncm-et-15 2021-08-20 16:07 01:14:21 1958176 500 433 1 66 +454.76 ± 42.55
150872 ncm-et-10 2021-08-20 14:57 01:16:07 1946608 500 431 0 69 +452.04 ± 41.5
150871 ncm-et-4 2021-08-20 14:54 01:16:34 1926130 500 409 3 88 +393.6 ± 36.68
150870 ncm-et-13 2021-08-20 14:54 01:14:31 1965687 500 421 1 78 +424.28 ± 38.99
150869 ncm-et-9 2021-08-20 14:54 01:15:55 1958635 500 422 0 78 +429.05 ± 38.92
150868 ncm-et-3 2021-08-20 14:53 01:14:12 1956496 500 420 0 80 +424.27 ± 38.4
150867 ncm-et-15 2021-08-20 14:52 01:14:50 1958471 500 417 0 83 +417.31 ± 37.67
150866 ncm-et-10 2021-08-20 13:40 01:16:43 1947250 500 420 1 79 +421.93 ± 38.73
150865 ncm-et-4 2021-08-20 13:38 01:15:23 1966152 500 418 0 82 +419.61 ± 37.91
150864 ncm-et-3 2021-08-20 13:38 01:15:00 1959389 500 429 1 70 +444.09 ± 41.26
150863 ncm-et-9 2021-08-20 13:38 01:15:20 1959390 500 427 1 72 +438.95 ± 40.66
150862 ncm-et-15 2021-08-20 13:37 01:14:09 1960768 500 419 0 81 +421.93 ± 38.15
150861 ncm-et-13 2021-08-20 13:37 01:16:30 1945081 500 427 1 72 +438.95 ± 40.66
150860 ncm-et-3 2021-08-20 12:23 01:14:36 1961075 500 423 4 73 +421.93 ± 40.4
150859 ncm-et-4 2021-08-20 12:23 01:15:01 1968324 500 423 1 76 +429.05 ± 39.52
150858 ncm-et-15 2021-08-20 12:23 01:13:53 1962461 500 421 0 79 +426.65 ± 38.66
150857 ncm-et-13 2021-08-20 12:23 01:13:47 1959549 500 426 0 74 +438.95 ± 40.01
150856 ncm-et-9 2021-08-20 12:23 01:14:40 1964460 500 413 2 85 +404.05 ± 37.32
150855 ncm-et-10 2021-08-20 12:22 01:16:28 1950435 500 420 2 78 +419.61 ± 39.03

Commit

Commit ID 18dcf1f09754284325157f2d270df10a09297958
Author Tomasz Sobczyk
Date 2021-08-20 06:50:25 UTC
Optimize and tidy up affine transform code. The new network caused some issues initially due to the very narrow neuron set between the first two FC layers. Necessary changes were hacked together to make it work. This patch is a mature approach to make the affine transform code faster, more readable, and easier to maintain should the layer sizes change again. The following changes were made: * ClippedReLU always produces a multiple of 32 outputs. This is about as good of a solution for AffineTransform's SIMD requirements as it can get without a bigger rewrite. * All self-contained simd helpers are moved to a separate file (simd.h). Inline asm is utilized to work around GCC's issues with code generation and register assignment. See https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101693, https://godbolt.org/z/da76fY1n7 * AffineTransform has 2 specializations. While it's more lines of code due to the boilerplate, the logic in both is significantly reduced, as these two are impossible to nicely combine into one. 1) The first specialization is for cases when there's >=128 inputs. It uses a different approach to perform the affine transform and can make full use of AVX512 without any edge cases. Furthermore, it has higher theoretical throughput because less loads are needed in the hot path, requiring only a fixed amount of instructions for horizontal additions at the end, which are amortized by the large number of inputs. 2) The second specialization is made to handle smaller layers where performance is still necessary but edge cases need to be handled. AVX512 implementation for this was ommited by mistake, a remnant from the temporary implementation for the new... This could be easily reintroduced if needed. A slightly more detailed description of both implementations is in the code. Overall it should be a minor speedup, as shown on fishtest: passed STC: LLR: 2.96 (-2.94,2.94) <-0.50,2.50> Total: 51520 W: 4074 L: 3888 D: 43558 Ptnml(0-2): 111, 3136, 19097, 3288, 128 and various tests shown in the pull request closes https://github.com/official-stockfish/Stockfish/pull/3663 No functional change
Copyright 2011–2024 Next Chess Move LLC