Разные матчи и турниры движков

[Владимир001]

Hardware: i9-10980XE CPU 36 cores, 64GB, gpu RTX 3080 + 2070

Развернуть для просмотра

Round Robin
LC0-version: v0.26.3
LC0-options:
"Threads": 3,
"MinibatchSize": 136,
"Cpuct": 2.147,
"CPuctAtRoot": 2.147
"NNCacheSize": 20000000,
"MaxCollisionEvents": 917,
"CPuctBase": 18368.000,
"CPuctFactor": 2.815,
"Backend":" multiplexing",
"BackendOptions":" (backend=cuda-fp16,gpu=0),(backend=cuda-fp16,gpu=1)",
"ShowHidden": true,
"Moves-left-max-effect": 0.2,
"MovesLeftThreshold": 0.0,
"MovesLeftSlope": 0.004,
"MovesLeftQuadraticFactor": 0.0,
"MovesLeftScaledFactor":1.0,
"MovesLeftConstantFactor": 0.0,
"StrictTiming": true,
"MaxPrefetch": 32,
"Temperature": 0.00,
"TempVisitOffset": 0.00,
"SmartPruningFactor": 1.33,
"MaxOutOfOrderEvalsFactor": 2.4,
"FpuStrategy": "reduction",
"FpuValue": 0.443,
"Ponder": false,
"PolicyTemperature": 1.607,
"MaxCollisionVisits": 9999,
"Moveoverhead": 20,
"RamLimitMb": 0,
"SyzygyFastPlay": true,
"SyzygyPath": "F:/syzygy"
Software: Octagon
Time control: 30 minutes + 30 seconds
Book: Sufi 19
Tablebases: 3/4/5/6 piece
Adjudication: -draw movenumber=35 movecount=10 score<=0.15 -resign movecount=10 score >=10
6 piece tablebase adjudication
Comment: Lc0 JH94-80 with default settings wins!

   # PLAYER              :  RATING  ERROR  POINTS  PLAYED   (%)  CFS(%)    W    D    L  D(%)
   1 Lc0-j94-80          :      55     42    35.5      60  59.2      73   14   43    3  71.7
   2 Stockfish 301120    :      35     40    33.5      60  55.8      78   12   43    5  71.7
   3 Lc0 T60-66390       :      10     40    31.0      60  51.7      83    9   44    7  73.3
   4 Dragon              :     -20     40    28.0      60  46.7      62    7   42   11  70.0
   5 Leelenstein 15      :     -30     39    27.0      60  45.0      74    3   48    9  80.0
   6 SF Classical        :     -50     41    25.0      60  41.7     ---    9   32   19  53.3

White advantage = 0.00 +/- 0.00
Draw rate (equal opponents) = 73.21 % +/- 3.66
MEDIANS
Name                    Games    Depth       NPS           Nodes
-------------------------------------------------------------------------------------
Lc0-j94-80          :      60     17           57022      5335868
Lc0 T60-66390       :      60     19           68098      7622882
Stockfish 301120    :      60     50        33327228   1406853600
Leelenstein 15      :      60     19          100520      9605020
Dragon              :      60     34        31324038   1452402867
SF Classical        :      60     44        42498878   1793085902

https://www.twitch.tv/navratil25

[свернуть]

[Владимир001]

Match: 66594 vs J94-80

Развернуть для просмотра

LC0 version: v0.26.3
LC0 options: Threads=2
Time control: N/A
Hardware: AMD Phenom II X6 1035T Processor, GTX 1660
Speed: 10k npm
Book: 10ply-100k
Book options: Opening order: Random. Play all openings for both white and black.
Tablebases: No Tablebases.
Adjudication: -draw movenumber=40 movecount=4 score=8 -resign movecount=4 score=500
Software: CuteChess-cli

  # PLAYER  :  RATING  ERROR  POINTS  PLAYED   (%)  CFS(%)    W     D    L  D(%)
  1 66594   :     0.0   ----   605.5    1200    50      81   96  1019   85    85
  2 J94-80  :    -3.3    7.4   594.5    1200    50     ---   85  1019   96    85

White advantage = 42.56 +/- 3.82
Draw rate (equal opponents) = 90.41 % +/- 1.13

[свернуть]

[Владимир001]

Match: 66555 vs SF-12 (16T), zz-tuned params
Hardware: RTX 3080, i9-10900kf
Time control: 120s + 2s

Развернуть для просмотра

Openings: 4mvs_eval080_089.pgn (Pohl's Human Unbalanced Openings-4moves, first 250 pos.)
Lc0 26.3 cuda-fp16 --show-hidden: nncache=2000000, threads=2, minibatch-size=112, move-overhead=0, mlh=on, smart-pruning-minimum-batches=80
MLH tcec-19: moves-left-max-effect=0.2, moves-left-threshold=0, moves-left-slope=0.004, moves-left-scaled-factor=1, moves-left-quadratic-factor=0, moves-left-constant-factor=0
ZZ-tuned params: root-has-own-cpuct-params=false, cpuct=1.745, cpuct-base=38739, cpuct-factor=3.894, fpu-value=0.330, policy-softmax-temp=1.359
Stockfish 12 bmi2: threads=16, hash=2048, move overhead=0
TBs and adj.: syzygy 5-men, draw 5 moves 5cp move 30, resign (two-sided) 5 moves 600cp
Comments: ZZ-tuned params on top, but CFS at 76%.
Games: https://gofile.io/d/sX3rao

# PLAYER                   : RATING  ERROR  PLAYED  (%)   CFS   W    D    L    D(%)
1 stockfish-051220-16T     :    0.0   ----   1000  51.30   72  230  566  204  56.60
2 lc0-263-66555-zznewtune  :   -6.0   20.4    500  49.30   76  107  279  114  55.80
3 lc0-263-66555            :  -16.3   18.0    500  48.10  ---   97  287  116  57.40

White advantage = 160.51 +/- 5.95
Draw rate (equal opponents) = 88.79 % +/- 3.49

Engine                 Depth   MIDG   EARLY   ENDG    LATE
lc0-263-66555          12.12  13.65 | 14.43 | 10.98 | 8.53
lc0-263-66555-zznewt   13.44  15.27 | 16.22 | 12.20 | 9.66
stockfish-051220-16T   37.43  31.85 | 32.63 | 39.42 | 52.44

[свернуть]

[Владимир001]

Match: Lc0 26.3 vs Stockfish 12 avx2 (4T)
Time control: 5s + 0.5s
Hardware: RTX 2060, Ryzen-9 4900H
Opening book: noob_3moves.pgn

Развернуть для просмотра

Stockfish: threads=4, hash=64, MoveOverhead=0
Lc0: cuda-fp16, threads=2, minibatch-size=60, mlh, move-overhead=0
mlh tcec-19: moves-left-max-effect=0.2, moves-left-threshold=0, moves-left-slope=0.004, moves-left-scaled-factor=1, moves-left-quadratic-factor=0, moves-left-constant-factor=0
TBs and adj.: syzygy 5-men, resign movecount=5 score=600, draw movenumber=30 movecount=5 score=5, time forfeit disabled
Comments: J94-80 added and takes the lead. J92-190 added.
Games: https://gofile.io/d/QXS62Q

# PLAYER                      :  RATING  ERROR  PLAYED    (%)   CFS    W      D     L    D(%)
1 Stockfish 12 4CPU           :     0.0   ----   18000  55.18   100  3337  13192  1471  73.29
2 lc0-263-j94-80              :   -26.5    7.9    2000  46.23    73   168   1513   319  75.65
3 lc0-263-cuda-j96-28         :   -30.1    8.0    2000  45.73    58   178   1473   349  73.65
4 lc0-263-cuda-66555          :   -31.1    8.1    2000  45.58    51   180   1463   357  73.15
5 lc0-263-cuda-j92-330        :   -31.3    7.7    2000  45.55    53   183   1456   361  72.80
6 lc0-263-cuda-j96-44         :   -31.6    8.0    2000  45.50    81   163   1494   343  74.70
7 lc0-263-cuda-66555-mlh-off  :   -36.6    8.5    2000  44.80    90   163   1466   371  73.30
8 lc0-263-cuda-66510-mlh-off  :   -43.9    7.5    2000  43.77    61   142   1467   391  73.35
9 lc0-263-cuda-J92-190        :   -45.5    8.3    2000  43.55    88   147   1448   405  72.40
10 lc0-263-cuda-66510         :   -51.9    7.4    2000  42.65   ---   147   1412   441  70.60

White advantage = 7.05 +/- 1.39
Draw rate (equal opponents) = 74.99 % +/- 0.31

[свернуть]

[Владимир001]

Match: T73 vs SF-12
Time control: 2s + 0.2s
Hardware: Ryzen-7 3750H, GTX 1660 Ti
Openings: TCEC SuFi 10-19 (497 pos.) + Balsa_v500 (500 pos.)

Развернуть для просмотра

Lc0 26.3 cudnn-fp16: nncache=200000, threads=1, minibatch-size=64, move-overhead=0
mlh tcec-19: moves-left-max-effect=0.2, moves-left-threshold=0, moves-left-slope=0.004, moves-left-scaled-factor=1, moves-left-quadratic-factor=0, moves-left-constant-factor=0
Stockfish 12: threads=1, hash=16, MoveOverhead=0
TBs and adj.: syzygy 5-men, draw 5 moves 5cp move 30, resign (two-sided) 5 moves 600cp, (time forfeit disabled)
Comments: 730843 added. A little setback for T73, though within the margin of error.
Games: https://gofile.io/d/fgv1t0

# PLAYER             : RATING  ERROR  PLAYED    (%)  CFS    W     D     L   D(%)
1 lc0-263-703810     :    1.8    9.8    1994  50.25   64   469  1066   459  53.46
2 stockfish-12-1cpu  :    0.0   ----   15952  51.70   89  4034  8427  3491  52.83
3 lc0-263-730730     :   -6.1    9.9    1994  49.15   70   435  1090   469  54.66
4 lc0-263-730843     :   -9.9    9.7    1994  48.62   67   432  1075   487  53.91
5 lc0-263-730620     :  -13.2   10.3    1994  48.17   70   462   997   535  50.00
6 lc0-263-730517     :  -16.8   10.3    1994  47.67   53   412  1077   505  54.01
7 lc0-263-730370     :  -17.3   10.9    1994  47.59   53   438  1022   534  51.25
8 lc0-263-730528     :  -17.9    9.2    1994  47.52   54   425  1045   524  52.41
9 lc0-263-730315     :  -18.6   11.9    1994  47.42  ---   418  1055   521  52.91

White advantage = 55.72 +/- 1.89
Draw rate (equal opponents) = 54.39 % +/- 0.42

[свернуть]

[Владимир001]

Vafra 3.0.0 vs Rybka 4.1; игра в шахматы с медленным контролем времени на ноутбуке HP Pavilion dv7, использующем Intel® Core ™ i7, с четырьмя гиперпоточными ядрами, оба движка думали только о своем собственном ходу, т.е. без размышлений; оба движка использовали шесть потоков, оба движка использовали одну и ту же вводную книгу, Fritz 16, минимум в 3 партиях до 8 ходов; У Вафры был доступ к таблицам эндшпиля на пять фигур , у Рыбки не было доступа к базам эндшпиля; Vafra 3.0.0 - это модифицированная Vafra 2.17.0 с NNUE, Vafra использовала nn-v0f000000070.nnue NNUE.

http://www.jurjevic.org.uk/chess/gam...Rybka_003.html

[Владимир001]

Arasan 22.1 wins 3 League JCER edition 41, 2020.12.15 - 2020.12.17

Temp 3' + 3" 180 games GUI-Fritz 17
Promotion to the 2 League: Arasan 22.1, Fritz 17.
Relegation to 4 League: Vajolet2 2.8.0, Marvin 4.0.1.

[Владимир001]

V3.0.0 (219) тест

NNUE Vafra 3.0.0 против не-NNUE Vafra (здесь называется Callida) 3.0.0; игра в шахматы с медленным контролем времени на ноутбуке HP Pavilion dv7, использующем Intel® Core ™ i7, с четырьмя гиперпоточными ядрами, оба движка думали только о своем собственном движении, т.е. оба движка использовали шесть потоков, оба движка использовали одну и ту же книгу открытия, Fritz 16, минимум в 3 партиях до 8 ходов; оба Вафраса имели доступ ко всем 5 и некоторым 6 фигурным эндшпильным базам ; Vafra 3.0.0 - это модифицированная Vafra 2.17.0 с поддержкой NNUE, NNUE Vafra использовала nn-v0f000000070.nnue NNUE.

http://www.jurjevic.org.uk/chess/gam...llida_005.html

[Владимир001]

https://lichess.org/study/EOddRjJ8

[Владимир001]

MaelicSuperStar wins JCER League group E2 edition 41, 2020.12.09 - 2020.12.12

Temp 3' + 3" 306 games
Promotion to Group E1: MaelicSuperStar, Sting 24.
Relegation to Group E3: RuyDos 1.1.11, Wyldchess 1.51, Coiled 0.6.

[Владимир001]

Hardware: i5-8400 CPU @ 2.80GHz, 6 cores, 32GB, gpu RTX 2060, 6GB
Gauntlet: SF-dev vs. LC0 384x30b nets

Развернуть для просмотра

Network IDs: 66512, 66592, 66637
LC0-version: v0.26.3
LC0-options: minibatch-size=90
SF-version: SF-dev 20120516
SF-options: 4 Threads, 512MB Hash
Software: Cutechess-CLI
Time control: LC0: 1,500 nodes/move, SF-dev: 500,000 nodes/move
Book: Chad's 8-ply book with 10000 positions, order=sequential
Tablebases: 6-men
Adjudication: -draw movenumber=50 movecount=5 score=8 -resign movecount=5 score=1000
Ordo: -a 0 -A "Stockfish.dev" -W -D -U "0,1,2,3,4,5,6,7,8,9,10" -s 2000 -n 1
Comment:

   # PLAYER           :  RATING  ERROR   POINTS  PLAYED   (%)  CFS(%)      W      D      L  D(%)
   1 lc0.net.66592    :    14.1    2.9  10394.5   20000    52      99   4200  12389   3411    62
-  2 lc0.net.66637    :     9.4    2.9  10263.0   20000    51      61   4083  12360   3557    62
   3 lc0.net.66512    :     8.9    3.0  10247.0   20000    51     100   3935  12624   3441    63
+  4 Stockfish.dev    :     0.0   ----  29095.5   60000    48     ---  10409  37373  12218    62

White advantage = 52.35 +/- 0.86
Draw rate (equal opponents) = 64.32 % +/- 0.20

PGN:
https://pixeldrain.com/u/s2rU69P8

[свернуть]

[Владимир001]

Beef 0.3.6 wins JCER League group E3 edition 41 2020.12.12 - 2020.12.17

Temp 3' + 3" 306 games GUI-Fritz 15
Promotion to Group E2: Beef 0.3.6, Stash 25.0.
Relegation to Group E4: Raven 1.10, CT800 1.41, Isa 2.0.83.

[Владимир001]

BrainLearn 12.1 wins JCER - Test new engines, 2020.12.17 - 2020.12.18

Temp 3' + 3" 112 games GUI-Fritz 17

[Владимир001]

Time control : 3 minutes
Computer : X 3450, 12 Gb, Windows 10
Engine setting : Default setting for both engines (nn-62ef826d1a6d.nnue)
Hashtable : 1024 mb
Threads : 4
Ponder : Off
GUI : ChessProgram 17
EGTB : 6 men
Games : 100
Book : W&B 3456 CTG for both engines

1. Cfish 171220 64 POPCNT NUMA +6/=92/-2 52.00% 52.0/100 14

2. SugaR AI 1.10 +2/=92/-6 48.00% 48.0/100 -14

Detail info>>>

[Владимир001]

Match: 66594 vs 66650
LC0 version: v0.26.3
LC0 options: Threads=2
Time control: N/A
Hardware: AMD Phenom II X6 1035T Processor, GTX 1660
Speed: 10k npm
Book: 10ply-100k
Book options: Opening order: Random. Play all openings for both white and black.
Tablebases: No Tablebases.
Adjudication: -draw movenumber=40 movecount=4 score=8 -resign movecount=4 score=500
Software: CuteChess-cli

  # PLAYER  :  RATING  ERROR  POINTS  PLAYED   (%)  CFS(%)    W     D    L  D(%)
  1 66594   :     0.0   ----   603.5    1200    50      71   94  1019   87    85
  2 66650   :    -2.1    7.4   596.5    1200    50     ---   87  1019   94    85

White advantage = 43.15 +/- 3.80
Draw rate (equal opponents) = 90.62 % +/- 1.13

[Владимир001]

66666 vs J98.3-8
LC0 version: v0.26.3
LC0 options: Defaults
Time control: N/A
Hardware: RBP-Server Farm
Speed: 100k npm
Book: openings-dd-24ply-15000.pgn
Book options: Opening order: Sequential. Play all openings for both white and black.
Tablebases: 6 man
Adjudication: -draw movenumber=40 movecount=4 score=8 -resign movecount=4 score=500
Software: CuteChess-GUI + Ordo v1.2.6

   # PLAYER     :  RATING  ERROR  POINTS  PLAYED   (%)     W     L     D  D(%)  CFS(%)
   1 66666      :    14.2    4.8  7803.0   15000    52  5393  4787  4820    32     100
   2 J98.3-8    :     0.0   ----  7197.0   15000    48  4787  5393  4820    32     ---

White advantage = 23.63 +/- 2.42
Draw rate (equal opponents) = 32.28 % +/- 0.38

[Владимир001]

Цитата: Сообщение от Владимир001

66666 vs J98.3-8
LC0 version: v0.26.3
LC0 options: Defaults
Time control: N/A
Hardware: RBP-Server Farm
Speed: 100k npm
Book: openings-dd-24ply-15000.pgn
Book options: Opening order: Sequential. Play all openings for both white and black.
Tablebases: 6 man
Adjudication: -draw movenumber=40 movecount=4 score=8 -resign movecount=4 score=500
Software: CuteChess-GUI + Ordo v1.2.6

   # PLAYER     :  RATING  ERROR  POINTS  PLAYED   (%)     W     L     D  D(%)  CFS(%)
   1 66666      :    14.2    4.8  7803.0   15000    52  5393  4787  4820    32     100
   2 J98.3-8    :     0.0   ----  7197.0   15000    48  4787  5393  4820    32     ---

White advantage = 23.63 +/- 2.42
Draw rate (equal opponents) = 32.28 % +/- 0.38

https://pixeldrain.com/u/ADywnS7s

[ChessMan]

Ryzen 3900X @4.2GHz
GUI: CuteChess 1.2
book: 8moves_v3.pgn
1 core, concurrency 8, hash 16 MB
Ponder=off
TC: 10+0.1

   # PLAYER                              :  RATING  ERROR  PLAYED   (%)    W      D    L  D(%)  CFS(%)
   1 Cfish_x64_171220_E_AVX2_mingw_10    :    3201      1   12392  50.3  966  10544  882  85.1      97
   2 Cfish++NNUE 171220 x64 AVX2         :    3199      1   12392  49.7  882  10544  966  85.1     ---

White advantage = 27.55 +/- 1.22
Draw rate (equal opponents) = 87.04 % +/- 0.33

PGN: https://pixeldrain.com/u/8ykVvTLW

[Владимир001]

OK, just a quick note, Vafra with flipped HalfKP 256x2-32-32 nn-v0f000000080.nnue, ML trained with one pass over d12-master multipv 4 data, manged to get a draw against Houdini 6, which is a good result for Vafra

V3.0.0 (219) test

Vafra 3.0.0 vs Houdini 6; a slow time control chess game, on HP Pavilion dv7 laptop, using Intel® Core™ i7, with four hyper-threaded Cores, both engines thought only on their own move, i.e., no pondering; both engines used six threads, both engines used the same opening book, Fritz 16, in min 3 games up to 8 moves; Vafra had access to all five- and some six-men endgame tablebases, Houdini had no access to endgame tablebases; Vafra 3.0.0 is a modified Vafra 2.17.0 with NNUE, Vafra was using nn-v0f000000080.nnue NNUE.

http://www.jurjevic.org.uk/chess/gam...udini_025.html

GOOD

[Владимир001]

Hardware: i5-8400 CPU @ 2.80GHz, 6 cores, 32GB, gpu RTX 2060, 6GB
Gauntlet: SF-dev vs. LC0 384x30b nets

Развернуть для просмотра

Network IDs: J98.3-8, 66512, 66592, 66637
LC0-version: v0.26.3
LC0-options: minibatch-size=90
SF-version: SF-dev 20120516
SF-options: 4 Threads, 512MB Hash
Software: Cutechess-CLI
Time control: LC0: 1,500 nodes/move, SF-dev: 500,000 nodes/move
Book: Chad's 8-ply book with 10000 positions, order=sequential
Tablebases: 6-men
Adjudication: -draw movenumber=50 movecount=5 score=8 -resign movecount=5 score=1000
Ordo: -a 0 -A "Stockfish.dev" -W -D -U "0,1,2,3,4,5,6,7,8,9,10" -s 2000 -n 1
Comment:

   # PLAYER             :  RATING  ERROR   POINTS  PLAYED   (%)  CFS(%)      W      D      L  D(%)
-  1 lc0.net.J98.3-8    :    14.8    3.0  10412.0   20000    52      62   4207  12410   3383    62
   2 lc0.net.66592      :    14.1    3.0  10394.5   20000    52      99   4200  12389   3411    62
   3 lc0.net.66637      :     9.4    3.0  10263.0   20000    51      61   4083  12360   3557    62
   4 lc0.net.66512      :     8.9    2.9  10247.0   20000    51     100   3935  12624   3441    63
+  5 Stockfish.dev      :     0.0   ----  38683.5   80000    48     ---  13792  49783  16425    62

White advantage = 52.33 +/- 0.75
Draw rate (equal opponents) = 64.27 % +/- 0.18

PGN:
https://pixeldrain.com/u/vV2Jf6Nn

[свернуть]

[Владимир001]

Match: J98.3-8 (value repair) Elo test
Lc0-version: 26.3
Lc0 options: SyzygyPath="D:/tb"
SF options: Threads=12 Hash=512 SyzygyPath="D:/tb"
Time/Node control: lc0 1k nodes per move, SF12dev 600k nodes per move
Hardware: RTX 2080 Ti, AMD 2950X
Books: book-ply12-unifen-Q-0.1-0.5.pgn, random with reverse.
Tablebases: 6-man Syzygy
Adjudication: 6-man Syzygy and -draw movenumber=50 movecount=6 score=10 -resign movecount=5 score=500 twosided=true
Chess Interface: cutechess-cli
Comments: J92-320 is the net transferred into T60 main training (and the parent net for J98.3 training). SF12.6 = stockfish_20112916.
Results:

   # PLAYER             :  RATING  ERROR   POINTS  PLAYED   (%)     W     L      D  D(%)  CFS(%)
   1 SF12.6             :     0.0   ----  20817.0   40000    52  9335  7701  22964    57     100
   2 lc0.net.J98.3-8    :   -11.8    3.2   9673.0   20000    48  3949  4603  11448    57     100
   3 lc0.net.J92-320    :   -17.7    3.1   9510.0   20000    48  3752  4732  11516    58     ---

(zipped game set too big)

[Владимир001]

Match: J98.1-16 (value repair) Elo test
Lc0-version: 26.3
Lc0 options: SyzygyPath="D:/tb"
SF options: Threads=12 Hash=512 SyzygyPath="D:/tb"
Time/Node control: lc0 1k nodes per move, SF12dev 600k nodes per move
Hardware: RTX 2080 Ti, AMD 2950X
Books: book-ply12-unifen-Q-0.1-0.5.pgn, random with reverse.
Tablebases: 6-man Syzygy
Adjudication: 6-man Syzygy and -draw movenumber=50 movecount=6 score=10 -resign movecount=5 score=500 twosided=true
Chess Interface: cutechess-cli
Comments: J92-320 is the net transferred into T60 main training (and the parent net for J98.1 training). SF12.6 = stockfish_20112916. The games for J92-320 are the same as for the J98.3 report above.
Results:

   # PLAYER              :  RATING  ERROR   POINTS  PLAYED   (%)     W     L      D  D(%)  CFS(%)
   1 SF12.6              :     0.0   ----  20788.5   40000    52  9333  7756  22911    57     100
   2 lc0.net.J98.1-16    :   -10.8    3.2   9701.5   20000    49  4004  4601  11395    57     100
   3 lc0.net.J92-320     :   -17.7    3.1   9510.0   20000    48  3752  4732  11516    58     ---

(zipped game set too big)

[Владимир001]

PGN - http://kirill-kryukov.com/chess/disc...e.php?id=47568

CCRL 40/15 Rating List - Custom engine selection
1199203 games played by 2756 programs, run by 23 testers
Ponder off, General books (up to 12 moves), 3-4-5 piece EGTB
Time control: Equivalent to 40 moves in 15 minutes on an Intel i7-4770k.
Computed on December 19, 2020 with Bayeselo based on 1'199'203 games
Tested by CCRL team, 2005-2020, http://ccrl.chessdom.com/ccrl/4040/

Rank                 Engine                   Elo   +    -   Score  AvOp  Games
1 ShashChess 15.0 64-bit 4CPU             3535  +28  -27  76.6% -177.7   438
  Stockfish 12 64-bit 4CPU                3517  +23  -23  74.6% -159.9   627
  SugaR NN 1.1 64-bit 4CPU                3474  +16  -16  70.9% -134.3  1270
  ShashChess 8.0 64-bit 4CPU              3445  +15  -15  65.8%  -94.5  1206
  ShashChess Pro 1.0 64-bit 4CPU          3391  +33  -33  65.6%  -94.0   256

[Владимир001]

PGN - http://kirill-kryukov.com/chess/disc...e.php?id=47539

Развернуть для просмотра

CCRL 40/15 Rating List - Custom engine selection
1197230 games played by 2747 programs, run by 23 testers
Ponder off, General books (up to 12 moves), 3-4-5 piece EGTB
Time control: Equivalent to 40 moves in 15 minutes on an Intel i7-4770k.
Computed on December 12, 2020 with Bayeselo based on 1'197'230 games
Tested by CCRL team, 2005-2020, http://ccrl.chessdom.com/ccrl/4040/

Rank                 Engine                   Elo   +    -   Score  AvOp  Games
1 Dragon by Komodo 64-bit                 3475  +21  -21  74.7% -166.8   756
  Komodo 13.3 64-bit                      3345  +13  -14  56.3%  -37.2  1586
  Komodo 14 64-bit                        3341  +13  -13  60.5%  -63.6  1731
  Komodo 13.1 64-bit                      3330  +17  -17  57.8%  -46.4  1062
  Komodo 11.3.1 64-bit                    3323  +14  -14  68.7% -123.3  1636
  Komodo 12.2.2 64-bit                    3323  +14  -14  59.8%  -63.7  1594
  Komodo 13.01 64-bit                     3321  +18  -18  59.0%  -52.6   832
  Komodo 11.2 64-bit                      3320  +14  -14  70.7% -139.6  1802
  Komodo 12.3 64-bit                      3320  +16  -16  63.0%  -83.9  1150
  Komodo 12 64-bit                        3315  +14  -15  67.0% -110.2  1509
  Komodo 10.4 64-bit                      3305  +14  -14  75.0% -166.8  1713
  Komodo 14 MCTS 64-bit                   3298  +19  -19  46.8%  +20.2   843
  Komodo 11.01 64-bit                     3292  +21  -20  71.3% -135.2   755
  Komodo 13.3 MCTS 64-bit                 3292  +19  -19  51.8%   -9.6   755
  Komodo 10.2 64-bit                      3290  +16  -16  73.3% -153.2  1318
  Komodo 13.01 MCTS 64-bit                3277  +17  -18  52.2%  -12.6   904
  Komodo 13.2.5 MCTS 64-bit               3277  +16  -16  44.9%  +33.1  1165
  Komodo 10.3 64-bit                      3269  +16  -16  71.2% -135.0  1248
  Komodo 10.1 64-bit                      3265  +19  -18  75.6% -165.1  1004
  Komodo 10 64-bit                        3260  +11  -10  59.0%  -57.2  3051
  Komodo 9.2 64-bit                       3258  +20  -20  73.5% -153.6   864
  Komodo 9.3 64-bit                       3253  +19  -19  73.3% -149.6   936
  Komodo 9.42 64-bit                      3241  +21  -21  71.2% -133.9   722
  Komodo 9 64-bit                         3235  +20  -20  71.8% -138.4   848
  Komodo 12.3 MCTS 64-bit                 3234  +14  -14  50.4%   -3.1  1616
  Komodo 9.1 64-bit                       3229  +21  -20  70.3% -129.9   770
  Komodo 12.2.2 MCTS 64-bit               3212  +16  -16  62.4%  -82.9  1350
  Komodo 8 64-bit                         3204  +16  -16  68.9% -117.5  1239
  Komodo 7a 64-bit                        3173  +20  -20  62.0%  -75.1   730
  Komodo TCEC 64-bit                      3143  +17  -17  60.1%  -61.7  1060
  Komodo 6 64-bit                         3141  +15  -15  60.3%  -62.9  1339
  Komodo CCT 64-bit                       3125  +15  -15  59.0%  -55.3  1383
  Komodo 5 64-bit                         3108  +11  -11  58.1%  -51.1  2800
  Komodo 5.1 64-bit                       3104  +17  -17  58.0%  -50.2   989
  Komodo 4 64-bit                         3098  +11  -11  57.5%  -50.5  2532
  Komodo 3 64-bit                         3090  +14  -14  58.8%  -58.6  1699
  Komodo 12.1.1 MCTS 64-bit               3079  +15  -15  46.9%  +24.6  1430
  Komodo 2.03 64-bit                      3045  +28  -28  50.1%   -1.5   369
  Komodo 2.01 64-bit                      3042  +21  -21  56.2%  -41.7   710
  Komodo 1.3 64-bit                       2972  +19  -19  46.6%  +20.7   891
  Komodo 1.2 64-bit                       2945  +16  -16  45.6%  +27.3  1246
  Komodo 1.0 64-bit                       2921  +14  -14  48.5%   +7.7  1706

Komodo-14.1 + Dragon for Windows.zip

[свернуть]

[Владимир001]

Match: 66594 vs 66662
LC0 version: v0.26.3
LC0 options: Threads=2
Time control: N/A
Hardware: AMD Phenom II X6 1035T Processor, GTX 1660
Speed: 10k npm
Book: 10ply-100k
Book options: Opening order: Random. Play all openings for both white and black.
Tablebases: No Tablebases.
Adjudication: -draw movenumber=40 movecount=4 score=8 -resign movecount=4 score=500
Software: CuteChess-cli
Comment: Third test of this kind in a row where 66594 comes out on top.

  # PLAYER  :  RATING  ERROR  POINTS  PLAYED   (%)  CFS(%)    W     D    L  D(%)
  1 66594  :     0.0   ----   609.0    1200    51      93   92  1034   74    86
  2 66662  :    -5.3    7.1   591.0    1200    49     ---   74  1034   92    86

White advantage = 35.76 +/- 3.70
Draw rate (equal opponents) = 90.24 % +/- 1.07

PGN:
https://pixeldrain.com/u/tuGe9TDT

[Владимир001]

просто обратите внимание, что nn-v0f000000090.nnue кажется нормальным; перевернут HalfKP 256x2-32-32 net weak_smile Вафра сумел отыграть ничью с Houdini 6 в существенно несбалансированной игре, обеспечив ничью в эндшпиле с разноцветными слонами с двумя оставленными пешками

V3.0.0 (219) тест

Vafra 3.0.0 против Houdini 6; игра в шахматы с медленным контролем времени на ноутбуке HP Pavilion dv7, использующем Intel® Core ™ i7, с четырьмя гиперпоточными ядрами, оба движка думали только о своем собственном ходу, т.е. без размышлений; оба движка использовали шесть потоков, оба движка использовали одну и ту же книгу открытия, Fritz 16, минимум в 3 партиях до 8 ходов; У Вафры был доступ ко всем таблицам эндшпиля на пять и несколько человек, у Гудини не было доступа к базам финальных игр; Vafra 3.0.0 - это модифицированная Vafra 2.17.0 с NNUE, Vafra использовала nn-v0f000000090.nnue NNUE.

http://www.jurjevic.org.uk/chess/gam...udini_026.html

[Владимир001]

Match: 66555 vs SF-011120 (16T), zz-tuned-params-101220
Hardware: RTX 3080, i9-10900kf
Time control: 120s + 2s
Openings: 250 pos. from TCEC SuFi's 10-19

Развернуть для просмотра

Lc0 26.3 cuda-fp16 --show-hidden: nncache=2000000, threads=2, minibatch-size=112, max-out-of-order-evals-factor=2, mlh=on, smart-pruning-minimum-batches=80
MLH tcec-19: moves-left-max-effect=0.2, moves-left-threshold=0, moves-left-slope=0.004, moves-left-scaled-factor=1, moves-left-quadratic-factor=0, moves-left-constant-factor=0
ZZ-tuned params: root-has-own-cpuct-params=false, cpuct=1.745, cpuct-base=38739, cpuct-factor=3.894, fpu-value=0.330, policy-softmax-temp=1.359
Stockfish 011120 bmi2: threads=16, hash=2048, Move Overhead=0
TBs and adj.: syzygy 5-men, draw 5 moves 5cp move 30, resign (two-sided) 5 moves 600cp
Comments: ZZ-tuned params again on top by a slight margin.
Games: https://gofile.io/d/oVoBEY

# PLAYER                  : RATING  ERROR  PLAYED  (%)   CFS   W    D    L   D(%)
1 stockfish-011120-16T    :    0.0   ----   1000  52.10   96  125  792  83  79.20
2 lc0-263-66555-zznewtune :  -12.2   13.9    500  48.30   74   43  397  60  79.40
3 lc0-263-66555           :  -17.9   12.1    500  47.50  ---   40  395  65  79.00

White advantage = 50.20 +/- 4.96
Draw rate (equal opponents) = 84.25 % +/- 1.28

Engine                 Depth   MIDG   EARLY    ENDG   LATE
lc0-263-66555          12.46  14.20 | 15.69 | 11.15 | 8.23
lc0-263-66555-zznewt   13.26  15.88 | 16.29 | 11.45 | 8.21
stockfish-011120-16T   38.17  32.68 | 32.12 | 39.99 | 53.02

[свернуть]

[Владимир001]

Match: Lc0 26.3 vs Stockfish 12 avx2 (4T)
Time control: 5s + 0.5s
Hardware: RTX 2060, Ryzen-9 4900H
Opening book: noob_3moves.pgn

Развернуть для просмотра

Stockfish: threads=4, hash=64, MoveOverhead=0
Lc0: cuda-fp16, threads=2, minibatch-size=60, mlh, move-overhead=0
mlh tcec-19: moves-left-max-effect=0.2, moves-left-threshold=0, moves-left-slope=0.004, moves-left-scaled-factor=1, moves-left-quadratic-factor=0, moves-left-constant-factor=0
TBs and adj.: syzygy 5-men, resign movecount=5 score=600, draw movenumber=30 movecount=5 score=5, time forfeit disabled
Comments: 66594 and 66727 added. T60 getting close to J94-80, presumably with a more entreprising style given the lower draw rate.
Games: https://gofile.io/d/gKI1B3

# PLAYER             : RATING  ERROR  PLAYED   (%)   CFS    W      D     L    D(%)
1 Stockfish 12 4CPU  :    0.0   ----   16000  54.56  100  2848  11764  1388  73.53
2 lc0-263-j94-80     :  -26.5    7.2    2000  46.23   55   168   1513   319  75.65
3 lc0-263-66727      :  -27.2    8.7    2000  46.13   67   196   1453   351  72.65
4 lc0-263-j96-28     :  -30.1    9.0    2000  45.73   57   178   1473   349  73.65
5 lc0-263-66555      :  -31.1    7.9    2000  45.58   51   180   1463   357  73.15
6 lc0-263-j92-330    :  -31.3    7.3    2000  45.55   53   183   1456   361  72.80
7 lc0-263-j96-44     :  -31.6    8.0    2000  45.50   62   163   1494   343  74.70
8 lc0-263-66594      :  -33.4    7.8    2000  45.25   98   173   1464   363  73.20
9 lc0-263-J92-190    :  -45.5    8.2    2000  43.55  ---   147   1448   405  72.40

[свернуть]

[Владимир001]

MrBob 0.9.0 wins JCER League E4 edition 41, 2020.12.19 - 2020.12.22

Temp 3' + 3" 306 games GUI-Fritz 15
Promotion to Group E3: MrBob 0.9.0, ECE 20.1.
Relegation to Group E5: Bumblebee 1.0, Dumb 1.4, Waxman 2017.

[Владимир001]

Match: Lc0 26.3 vs Stockfish 201218 bmi2 (8T)
Time control: 3s + 0.3s
Hardware: RTX 3070, i7-10700
Openings: TCEC SuFi's 10-19 (500 pos.) with a few changes (about 10 pos. to avoid doublons).

Развернуть для просмотра

Stockfish: threads=8, hash=256, MoveOverhead=0
Lc0: cuda-fp16, threads=1, minibatch-size=80, mlh, move-overhead=0
mlh tcec-19: moves-left-max-effect=0.2, moves-left-threshold=0, moves-left-slope=0.004, moves-left-scaled-factor=1, moves-left-quadratic-factor=0, moves-left-constant-factor=0
TBs and adj.: syzygy 5-men, resign movecount=5 score=600, draw movenumber=30 movecount=5 score=5, time forfeit disabled
Comments: J94-80 still on top, 66740 not far behind.
Games: https://gofile.io/d/VzELLk

# PLAYER              : RATING ERROR PLAYED  (%)  CFS   W    D    L    D(%)  Depth
1 lc0-263-j94-80      :   10.4  12.4  1000  51.40  56  214  600  186  60.00   8.36
2 lc0-263-66740       :    8.9  13.8  1000  51.20  74  221  582  197  58.20   8.46
3 lc0-263-66662       :    2.6  12.9  1000  50.35  65  209  589  202  58.90   8.47
4 stockfish-201218-8T :    0.0  ----  4000  49.35  65  788 2372  840  59.30  25.36
5 lc0-263-j92-280     :   -2.6  13.4  1000  49.65 ---  196  601  203  60.10   8.43

White advantage = 83.41 +/- 3.32
Draw rate (equal opponents) = 64.05 % +/- 0.74

[свернуть]