Разные матчи и турниры движков

[Владимир001]

Raubfisch X47d2b and Fisherov 0.98h wins CEDR - Acer New Engines (for older hardware), 2021.11.07 - 2021.11.12

Time 15'+10" Acer i3-3217U with 3,9 GB GUI Arena 3.5.1
Book Perfect 2021 CEDR 6.079 games 01.11.2021 Table: Scid vs PC
CEDR Test Tournament: 182 games

[Владимир001]

Stockfish 14.1 wins CEDR - New Engines Tournament, 2021.11.07 - 2021.11.12

Time 15'+10" Lenovo Z710 i7-4710MG 8GB RAM GUI Fritz 17
Book Perfect 2021 CEDR 6.079 games 01.11.2021 Table: Scid vs PC
CEDR Test Tournament: 182 games

[Владимир001]

Gauntlet: 753723 vs. 753521, 761065
LC0 version: v0.28.0 cuda
LC0 options: multi_stream=true, mbs=48
Hardware: i9-7960x CPU, RTX 3060
Time control: LC0: 1k npm
Book: book-ply8-unifen-Q-0.0-0.25, order=sequential
Tablebases: none
Adjudication: -6man tb's -draw movenumber=50 movecount=5 score=8 -resign movecount=5 score=6500
Software: cutechess-cli
Ordo: -a 0 -U "0,1,2,3,4,5,6,7,8,9,10" -A "lc0.net.753723" -W -D -n16 -s10000 -C
Comment:

   # PLAYER            :  RATING  ERROR  POINTS  PLAYED   (%)  CFS(%)    W     D    L  D(%)
   1 lc0.net.753521    :     0.4    8.3  1001.0    2000    50      53  282  1438  280    72
   2 lc0.net.753723    :     0.0   ----  2044.5    4000    51     100  631  2827  542    71
   3 lc0.net.761065    :   -16.2    8.3   954.5    2000    48     ---  260  1389  351    69

White advantage = 39.91 +/- 3.00
Draw rate (equal opponents) = 72.45 % +/- 0.73

https://pixeldrain.com/u/tNpbevpi

[Владимир001]

Hardware: AMD Ryzen 9 5900X CPU @ 3.70GHz, 12 cores, 64GB, gpu RTX 3080, 10GB
Gauntlet: SF-dev vs. LC0 384x30b nets

Развернуть для просмотра

Network IDs: J94-100, 607646, 607831, 607892, 607925, 607971, 607996, 608059, 608112, 608169, 608207, 608242, 608285, 608326, 608363, 608405, 608446, 608497, 608551, 608604, 608653, 608703, 608744, 608795, 608856, 608904, 608960, 690013, 609051, 609085, 609119, 609146, 609169, 609188, 609202, 609288, 609288, 609471, 609524, 609626, 609677, 609722, 609833, 609841, 609953, 609958, 609969, 609973, 609996, 610016, 610024, 610031, 610034, 610035, 610038, 610044, 610053, 610093, 610110, 610161, 610221, 610357, 610363, 610369
LC0-version: v0.27.0
LC0-options: minibatch-size=112, MultiGather=true
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: The aim of the test is to order T60 nets. The order is considered sure with 99% CFS (a bit more than 5 Elo). CFS matrix and detailed results in 7z file.

[свернуть]

   # PLAYER             :  RATING  ERROR    POINTS   PLAYED   (%)  CFS(%)       W       D       L  D(%)
   1 lc0.net.610024     :    46.8    3.0   11294.5    20000    56      76    5076   12437    2487    62
-  2 lc0.net.609958     :    45.3    3.0   11254.0    20000    56      54    5040   12428    2532    62
...
+ 21 lc0.net.610369     :    36.9    3.0   11023.5    20000    55      55    4785   12477    2738    62
...
- 64 Stockfish.dev      :     0.0   ----  570540.5  1260000    45     ---  180394  780293  299313    62

White advantage = 55.31 +/- 0.19
Draw rate (equal opponents) = 64.90 % +/- 0.05

PGN: https://pixeldrain.com/u/No5msZGf

[Владимир001]

Network IDs: 609722, 609953, 609958, 609969, 610024, 610034, 610044, 610053, 610093, 610110, 610161, 610221, 610357, 610363, 610369
Book: book-ply8-unifen-Q-0.0-0.25.pgn and book-ply8-unifen-Q-0.25-0.40.pgn (10K openings in both books), order=sequential
Comment: Adding 40K games to tests with Chad's book. The aim of the test is to order T60 nets. The order is considered sure with 99% CFS (a bit more than 3.5 Elo). CFS matrix and detailed results in 7z file.

   # PLAYER            :  RATING  ERROR    POINTS  PLAYED   (%)  CFS(%)       W       D       L  D(%)
-  1 lc0.net.609958    :    47.9    1.8   33897.0   60000    56      87   17124   33546    9330    56
   2 lc0.net.609969    :    46.4    1.8   33778.0   60000    56      77   16922   33712    9366    56
   3 lc0.net.610024    :    45.4    1.8   33698.0   60000    56      57   16894   33608    9498    56
...
  12 lc0.net.609722    :    41.1    1.9   33345.5   60000    56      61   16764   33163   10073    55
  13 lc0.net.610363    :    40.7    1.8   33316.0   60000    56      81   16499   33634    9867    56
  14 lc0.net.610357    :    39.6    1.9   33225.0   60000    55      61   16301   33848    9851    56
+ 15 lc0.net.610369    :    39.2    1.9   33194.5   60000    55     100   16494   33401   10105    56
- 16 Stockfish.dev     :     0.0   ----  397087.5  900000    44     ---  145167  503841  250992    56

White advantage = 74.82 +/- 0.24
Draw rate (equal opponents) = 60.14 % +/- 0.06

PGN: https://pixeldrain.com/l/vsuDkaM2#item=0

[Владимир001]

Fisherov 0.98h wins CEDR - Fritz 17 Tournament, 2021.11.10 - 2021.11.13

Time 15'+10" HP Pavilion i5-1035G1 8GB RAM GUI Fritz 17
Book Perfect 2021 CEDR 6.079 games 01.11.2021 Table: Scid vs PC
CEDR Test Tournament: 90 games

[Владимир001]

Crystal 201021 wins CEDR - Android Chess Engines Tournament, 2021.11.14

Time: 3' 12 games download, GUI - Chess Tournament,
Tablet - Smasung Galaxy TAB S6 Lite, 4GB RAM, all 8.855 games

[Владимир001]

Stockfish dev (Tweak initial aspiration window. Fri Nov 5 22:22:30 2021 +0100) vs Stockfish 8.
36x time odds match: 5+0.1 (SF dev) ;180+0.1 (SF 8) (time in seconds)
UHO fishtest default book.
SF dev with default settings + full 6 men TB.
SF 8 with default settings except 1 GB of RAM and full 6 men TB.

# PLAYER           :  RATING  ERROR  POINTS  PLAYED   (%)    W    D    L  D(%)  OppN  CFS(%)
   1 Stockfish dev    :    3185     32    70.5     100  70.5   52   37   11  37.0     1     100
   2 Stockfish 8      :    3000     32    29.5     100  29.5   11   37   52  37.0     1     ---

White advantage = 164.34 +/- 34.53
Draw rate (equal opponents) = 52.67 % +/- 6.65

[Владимир001]

Vafra 3.3.0 с nn-v0f000007001.nnue NNUE против Leela 0.28.0 с 744204.pb.gz NN в 20 специальных тестах блиц-игр 5 м + 5 с, Vafra использовала все 5 и некоторые 6 EGTB, у Лилы не было доступа к EGTB, версия процессора Leela vas, тестовый запуск на ноутбуке Intel 4 Core HP с Windows 10

Личная статистика:

1) Lc0 v0.28.0           289.0 :     20 (+14,=5,-1),  82.5 %

   vs.                          :  games (  +, =, -),   (%) :    Diff,    SD, CFS (%)
   Vafra 3.3.0 x64 SSE41        :     20 ( 14, 5, 1),  82.5 :  +289.0,  92.0,   99.9

2) Vafra 3.3.0 x64 SSE41   0.0 :     20 (+1,=5,-14),  17.5 %

   vs.                          :  games ( +, =,  -),   (%) :    Diff,    SD, CFS (%)
   Lc0 v0.28.0                  :     20 ( 1, 5, 14),  17.5 :  -289.0,  92.0,    0.1

[Владимир001]

Match: 10b Attention Policy vs. 10b Baseline

Развернуть для просмотра

Lc0 version: v0.29.0-dev
Lc0 options: --backend=onnx-cuda --minibatch-size=32 --threads=1
Hardware: RTX 2070 super
Time Control: inf, 1k and 3k npm
Book: 1k npm: book-ply4-unifen-Q-0.0-0.25; 3k npm: book-ply8-unifen-Q-0.25-0.40, book-ply6-unifen-Q-0.0-0.25, book_1move_FRC_cp1-24_27736pos, book_1move_FRC_cp25-49_37303pos
Tablebases: no
Adjudication: 5-man TB, draw after 250 moves. resign movecount=3 score=700 (stepped down to score=650 for unbalanced FRC)
Software: cutechess-cli
Ordo: -Q -N 0 -D -a 0 -W -n4 -s5000 -U "0,1,2,3,4,5,6,7,8,9,10" -A baseline_10b_800k
Comment: Here are the results of all of the >1 npm matches I have run between the 10b attention and baseline nets so far. 1000 openings, 2000 games each. Results are pretty consistent across all tests, including the step up from 1k to 3k npm. Will also mention some of dje’s testing at higher npm with Ceres in #test-discuss .

[свернуть]

1k npm 8-ply balanced:
   # PLAYER                  :  RATING  ERROR  POINTS  PLAYED   (%)  CFS(%)    W     D    L  D(%)
   1 attention_10b_800k      :   15.04  10.16  1042.5    2000  52.1     100  465  1155  380  57.8
   2 baseline_10b_800k       :    0.00   ----   957.5    2000  47.9     ---  380  1155  465  57.8

White advantage = 33.64 +/- 5.13
Draw rate (equal opponents) = 58.52 % +/- 1.12

3k npm 8-ply unbalanced (25-40cp):
   # PLAYER                :  RATING  ERROR  POINTS  PLAYED   (%)  CFS(%)    W    D    L  D(%)
   1 attention_10b_800k    :   13.00  11.04  1033.5    2000  51.7      99  537  993  470  49.6
   2 baseline_10b_800k     :    0.00   ----   966.5    2000  48.3     ---  470  993  537  49.6

White advantage = 112.69 +/- 5.61
Draw rate (equal opponents) = 55.69 % +/- 1.26

3k npm 6-ply balanced:
   # PLAYER                :  RATING  ERROR  POINTS  PLAYED   (%)  CFS(%)    W     D    L  D(%)
   1 attention_10b_800k    :   14.83   8.62  1042.0    2000  52.1     100  357  1370  273  68.5
   2 baseline_10b_800k     :    0.00   ----   958.0    2000  47.9     ---  273  1370  357  68.5

White advantage = 29.21 +/- 4.61
Draw rate (equal opponents) = 69.47 % +/- 1.01

3k npm FRC balanced:
   # PLAYER                :  RATING  ERROR  POINTS  PLAYED   (%)  CFS(%)    W     D    L  D(%)
   1 attention_10b_800k    :   12.11   9.59  1034.5    2000  51.7      99  417  1235  348  61.8
   2 baseline_10b_800k     :    0.00   ----   965.5    2000  48.3     ---  348  1235  417  61.8

White advantage = 12.11 +/- 4.91
Draw rate (equal opponents) = 61.94 % +/- 1.10

3k npm FRC unbalanced (25-49cp):
   # PLAYER                :  RATING  ERROR  POINTS  PLAYED   (%)  CFS(%)    W     D    L  D(%)
   1 attention_10b_800k    :   16.33  10.06  1046.0    2000  52.3     100  461  1170  369  58.5
   2 baseline_10b_800k     :    0.00   ----   954.0    2000  47.7     ---  369  1170  461  58.5

White advantage = 38.79 +/- 5.07
Draw rate (equal opponents) = 59.54 % +/- 1.12

https://pixeldrain.com/u/Qp4TeJt4

[Владимир001]

Hardware: AMD Ryzen 9 5900X CPU @ 3.70GHz, 12 cores, 64GB, gpu RTX 3080, 10GB
Gauntlet: SF-dev vs. LC0 384x30b nets

Развернуть для просмотра

Network IDs: J94-100, 607646, 607831, 607892, 607925, 607971, 607996, 608059, 608112, 608169, 608207, 608242, 608285, 608326, 608363, 608405, 608446, 608497, 608551, 608604, 608653, 608703, 608744, 608795, 608856, 608904, 608960, 690013, 609051, 609085, 609119, 609146, 609169, 609188, 609202, 609288, 609288, 609471, 609524, 609626, 609677, 609722, 609833, 609841, 609953, 609958, 609969, 609973, 609996, 610016, 610024, 610031, 610034, 610035, 610038, 610044, 610053, 610093, 610110, 610161, 610221, 610357, 610363, 610366, 610369
LC0-version: v0.27.0
LC0-options: minibatch-size=112, MultiGather=true
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: The aim of the test is to order T60 nets. The order is considered sure with 99% CFS (a bit more than 5 Elo). CFS matrix and detailed results in 7z file.

[свернуть]

   # PLAYER             :  RATING  ERROR    POINTS   PLAYED   (%)  CFS(%)       W       D       L  D(%)
   1 lc0.net.610024     :    46.8    3.1   11294.5    20000    56      75    5076   12437    2487    62
-  2 lc0.net.609958     :    45.3    3.0   11254.0    20000    56      54    5040   12428    2532    62
...
+ 18 lc0.net.610366     :    38.7    2.9   11074.0    20000    55      60    4828   12492    2680    62
...
- 65 Stockfish.dev      :     0.0   ----  579466.5  1280000    45     ---  183074  792785  304141    62

White advantage = 55.28 +/- 0.19
Draw rate (equal opponents) = 64.91 % +/- 0.05

PGN: https://pixeldrain.com/u/mu75STNz

[Владимир001]

Network IDs: 609722, 609953, 609958, 609969, 610024, 610034, 610044, 610053, 610093, 610110, 610161, 610221, 610357, 610363, 610366, 610369
Book: book-ply8-unifen-Q-0.0-0.25.pgn and book-ply8-unifen-Q-0.25-0.40.pgn (10K openings in both books), order=sequential
Comment: Adding 40K games to tests with Chad's book. The aim of the test is to order T60 nets. The order is considered sure with 99% CFS (a bit more than 3.5 Elo). CFS matrix and detailed results in 7z file.

   # PLAYER            :  RATING  ERROR    POINTS  PLAYED   (%)  CFS(%)       W       D       L  D(%)
-  1 lc0.net.609958    :    47.9    1.9   33897.0   60000    56      86   17124   33546    9330    56
   2 lc0.net.609969    :    46.4    1.9   33778.0   60000    56      77   16922   33712    9366    56
   3 lc0.net.610024    :    45.4    1.8   33698.0   60000    56      57   16894   33608    9498    56
...
  12 lc0.net.609722    :    41.1    1.8   33345.5   60000    56      61   16764   33163   10073    55
  13 lc0.net.610363    :    40.7    1.8   33316.0   60000    56      66   16499   33634    9867    56
+ 14 lc0.net.610366    :    40.1    1.8   33271.0   60000    55      67   16453   33636    9911    56
  15 lc0.net.610357    :    39.6    1.8   33225.0   60000    55      61   16301   33848    9851    56
  16 lc0.net.610369    :    39.2    1.9   33194.5   60000    55     100   16494   33401   10105    56
- 17 Stockfish.dev     :     0.0   ----  423816.5  960000    44     ---  155078  537477  267445    56

White advantage = 74.80 +/- 0.24
Draw rate (equal opponents) = 60.14 % +/- 0.05

PGN: https://pixeldrain.com/l/rCx78VpU#item=0

[Владимир001]

Vafra 3.3.0 with nn-v0f000007002.nnue NNUE vs Leela 0.28.0 with 744204.pb.gz NN in 20 5m+5s blitz games ad hoc test, Vafra was using all 5- and some 6-men EGTB, Leela all 5-men EGTB, Leela was CPU version, test run on Intel 4 Core HP laptop with Windows 10; first 8 games without code change in Vafra, the remaining 12 games with the EGTB probing code change

Head to head statistics:

1) Lc0 v0.28.0           150.5 :     20 (+10,=8,-2),  70.0 %

   vs.                          :  games (  +, =, -),   (%) :    Diff,    SD, CFS (%)
   Vafra 3.3.0 x64 SSE41        :     20 ( 10, 8, 2),  70.0 :  +150.5,  68.8,   98.6

2) Vafra 3.3.0 x64 SSE41   0.0 :     20 (+2,=8,-10),  30.0 %

   vs.                          :  games ( +, =,  -),   (%) :    Diff,    SD, CFS (%)
   Lc0 v0.28.0                  :     20 ( 2, 8, 10),  30.0 :  -150.5,  68.8,    1.4

[Владимир001]

Fischer vs Karpov games

https://www.chess2u.com/t17671-fisch...v-games#110835

[Владимир001]

SF14.1 vs SF8
100x time odds match: 4+0.04 (SF14.1), 400+4 (SF8)
Perfect 2021 (What Mark Young uses for his streams)
Default settings for SF14.1
SF8 has 256 mb hash

Score of stockfish_14.1 vs stockfish_8: 13 - 14 - 73 [0.495]
...      stockfish_14.1 playing White: 7 - 4 - 39  [0.530] 50
...      stockfish_14.1 playing Black: 6 - 10 - 34  [0.460] 50
...      White vs Black: 17 - 10 - 73  [0.535] 100
Elo difference: -3.5 +/- 35.5, LOS: 42.4 %, DrawRatio: 73.0 %
100 of 100 games finished.

[Владимир001]

89TH AMATEUR SERIES (Division 4)

Intel i7 Quad
ChessGUI
256mb hash each
3-4-5 piece tablebases
Ponder off
IECGMasters.cgb book
40 moves in 16 minutes repeating (adapted for the CCRL)
4 cycles 44 rounds
All engines 64-bit 1CPU (where available)

Цитата:

The top engine will gain automatic promotion and the bottom engine will gain automatic relegation. Any other changes of division will be at my discretion and may not necessarily involve playoff matches. This will definitely be the final series that Vajolet is eligible for until a new version is released.

Final Standings

29.5 - Vajolet2 2.8 64-bit
28.0 - chess22k 1.14 64-bit
26.0 - Pirarucu 3.3.5 64-bit
22.0 - Monolith 2.0 64-bit
21.5 - Drofa 3.2.0 64-bit
21.0 - Cheng 4.41 64-bit
21.0 - Bagatur 2.2g 64-bit
20.0 - Topple 0.8.1 64-bit
19.5 - Velvet 3.0.0 64-bit
19.0 - Rodent IV 0.30 64-bit
18.5 - Mr Bob 1.0.0 64-bit
18.0 - FabChess 1.16 64-bit


The complete tournament pgn (zipped) can be downloaded here:
http://kirill-kryukov.com/chess/disc...e.php?id=49426

[Владимир001]

Fisherov 0.98b4 wins CEDR - Android Chess Engines Tournament, 2021.11.14-15

Time: 3' 132 games download, GUI - Chess Tournament,
Tablet - Smasung Galaxy TAB S6 Lite, 4GB RAM, all 8.855 games

[Владимир001]

Vafra 3.3.0 with nn-v0f000007002.nnue NNUE vs Stockfish 14.1 in 20 5m+5s blitz games ad hoc test, Vafra was using all 5- and some 6-men EGTB, Stockfish had no access to EGTB, test run on Intel 4 Core HP laptop with Windows 10

Head to head statistics:

1) Stockfish 14.1        447.0 :     20 (+17,=3,-0),  92.5 %

   vs.                          :  games (  +, =, -),   (%) :    Diff
   Vafra 3.3.0 x64 SSE41        :     20 ( 17, 3, 0),  92.5 :  +447.0

2) Vafra 3.3.0 x64 SSE41   0.0 :     20 (+0,=3,-17),   7.5 %

   vs.                          :  games ( +, =,  -),   (%) :    Diff
   Stockfish 14.1               :     20 ( 0, 3, 17),   7.5 :  -447.0

[Владимир001]

Fisherov 0.98h and Raubfisch X47d2c wins CEDR - Acer New Engines Tournament (for older hardware), 2021.11.12 - 2021.11.16

Time 15'+10" Acer i3-3217U with 3,9 GB Memory GUI Arena 3.5.1
Book Perfect 2021 CEDR 6.079 games 01.11.2021 Table: Scid vs PC
CEDR Test Tournament: 132 games

[Владимир001]

Comment: Games nearly doubled from https://discord.com/channels/4254194...18742631727134 , see Book:.
Same place order, 10+ worse nets performance.
Lc0 version: 0.28.0 (Linux compiled Cuda 11.3)
Lc0 options: cuda-fp16 multi_stream=true MoveOverheadsMs=50
Stockfish options: Threads=16 Hash=2048 Move Overhead=10 Sy.50MR=false
Hardware: Ryzen 7 2700X 8c16t | RTX 2060 (both stock speed)
Time control: Lc0 G/15 secs + 0.25 secs , Stockfish G/5 secs + 0.1 sec
Book: Each net 1st 331 openings (662 games) added to prior zz-2_All from 8mvsbig+80_+109.epd https://github.com/official-stockfish/books
Tablebases: Syzygy-6
Adjudication: Draw: Move 35, count 2, 0.12; Resign: count 2, score 7.00, two-sided
Software: cutechess-cli, Ordo 1.2.6 -s 14000 -n 12

   # PLAYER                  :  RATING  ERROR  POINTS  PLAYED   (%)  CFS(%)     W     D    L  D(%)
   1 Stockfish_21110813      :     0.0   ----  4377.0    7000    63     100  2551  3652  797    52
   2 Lc0-0.28.0_609958-5c    :   -89.7   12.3   543.0    1400    39      77   185   716  499    51
   3 Lc0-0.28.0_610493       :   -96.2   12.3   532.0    1400    38      64   166   732  502    52
   4 Lc0-0.28.0_610235-6P    :   -99.5   12.4   526.5    1400    38      79   151   751  498    54
   5 Lc0-0.28.0_J92-300      :  -106.7   12.4   514.5    1400    37      69   156   717  527    51
   6 Lc0-0.28.0_610461       :  -111.2   12.5   507.0    1400    36     ---   139   736  525    53

Original zz-2__All.pgn merged with new games in the same file:
https://pixeldrain.com/u/n29QavzE

[Владимир001]

Fisherov 0.98h wins CEDR Fritz Tournament, 12-17.11.2021

Time 15'+10" HP Pavilion, i5-1035G1, 8GB RAM GUI Fritz 17
Book --- CEDR 6.079 games 01.11.2021 Table: Scid vs PC
CEDR Test Tournament: 182 games

[Владимир001]

CONGRATULATIONS TO THE WINNER OF THIS TOURNAMENT!!!
THANKING TO ALL BOOK CREATORS FOR CREATING AMAZING AND STRONG BOOKS!

Settings Used for the Tournament:
GUI Used= Fritz 15
Engine Used= SF PB 14.1(Thank You To Massimiliano Goi [My Friend], Thank You For The Authors Too)
Time Control= 2mins+1sec increment
Threads= 3per engine
Hash= 2048mb

GAMES PLAYED IN THIS BOOK TOURNAMENT:
Link1= https://pixeldrain.com/u/gLVDpuhR
Link2= https://www.mediafire.com/file/74sgr...Tourn.rar/file

[Владимир001]

Tests results
Test suite:
https://github.com/amchess/ShashChes...-suite-results
Match:
https://github.com/amchess/ShashChess/wiki/Matches

[Владимир001]

Hardware: AMD Ryzen 9 5900X CPU @ 3.70GHz, 12 cores, 64GB, gpu RTX 3080, 10GB
Gauntlet: SF-dev vs. LC0 384x30b nets

Развернуть для просмотра

Network IDs: J94-100, 607646, 607831, 607892, 607925, 607971, 607996, 608059, 608112, 608169, 608207, 608242, 608285, 608326, 608363, 608405, 608446, 608497, 608551, 608604, 608653, 608703, 608744, 608795, 608856, 608904, 608960, 690013, 609051, 609085, 609119, 609146, 609169, 609188, 609202, 609288, 609288, 609471, 609524, 609626, 609677, 609722, 609833, 609841, 609953, 609958, 609969, 609973, 609996, 610016, 610024, 610031, 610034, 610035, 610038, 610044, 610053, 610093, 610110, 610161, 610221, 610357, 610363, 610366, 610369, 610428
LC0-version: v0.27.0
LC0-options: minibatch-size=112, MultiGather=true
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: The aim of the test is to order T60 nets. The order is considered sure with 99% CFS (a bit more than 5 Elo). CFS matrix and detailed results in 7z file.

[свернуть]

   # PLAYER             :  RATING  ERROR    POINTS   PLAYED   (%)  CFS(%)       W       D       L  D(%)
   1 lc0.net.610024     :    46.8    3.0   11294.5    20000    56      75    5076   12437    2487    62
-  2 lc0.net.609958     :    45.3    3.0   11254.0    20000    56      54    5040   12428    2532    62
...
+ 16 lc0.net.610428     :    39.8    3.0   11103.5    20000    56      59    4831   12545    2624    63
...
- 66 Stockfish.dev      :     0.0   ----  588363.0  1300000    45     ---  185698  805330  308972    62

White advantage = 55.27 +/- 0.19
Draw rate (equal opponents) = 64.93 % +/- 0.05

PGN: https://pixeldrain.com/u/yPNWm6hZ

[Владимир001]

Network IDs: 609722, 609953, 609958, 609969, 610024, 610034, 610044, 610053, 610093, 610110, 610161, 610221, 610357, 610363, 610366, 610369, 610428
Book: book-ply8-unifen-Q-0.0-0.25.pgn and book-ply8-unifen-Q-0.25-0.40.pgn (10K openings in both books), order=sequential
Comment: Adding 40K games to tests with Chad's book. The aim of the test is to order T60 nets. The order is considered sure with 99% CFS (a bit more than 3.5 Elo). CFS matrix and detailed results in 7z file.

# PLAYER            :  RATING  ERROR    POINTS   PLAYED   (%)  CFS(%)       W       D       L  D(%)
   1 lc0.net.609958    :    47.9    1.8   33897.0    60000    56      86   17124   33546    9330    56
   2 lc0.net.609969    :    46.4    1.9   33778.0    60000    56      77   16922   33712    9366    56
   3 lc0.net.610024    :    45.4    1.8   33698.0    60000    56      57   16894   33608    9498    56
   4 lc0.net.610093
   5 lc0.net.610053
   6 lc0.net.610044
   7 lc0.net.610034
   8 lc0.net.609953
-  9 lc0.net.610110
- 10 lc0.net.610161
- 11 lc0.net.610221
  12 lc0.net.609722    :    41.0    1.9   33345.5    60000    56      61   16764   33163   10073    55
- 13 lc0.net.610363    :    40.7    1.8   33316.0    60000    56      60   16499   33634    9867    56
+ 14 lc0.net.610428    :    40.4    1.8   33289.0    60000    55      57   16341   33896    9763    56
- 15 lc0.net.610366    :    40.1    1.9   33271.0    60000    55      66   16453   33636    9911    56
- 16 lc0.net.610357    :    39.6    1.9   33225.0    60000    55      61   16301   33848    9851    56
- 17 lc0.net.610369    :    39.2    1.9   33194.5    60000    55     100   16494   33401   10105    56
  18 Stockfish.dev     :     0.0   ----  450527.5  1020000    44     ---  164841  571373  283786    56

White advantage = 74.69 +/- 0.23
Draw rate (equal opponents) = 60.15 % +/- 0.05

PGN: https://pixeldrain.com/l/RDYTRTXZ#item=0

[Владимир001]

(  Games 150/180)
Median Speeds
Name                    Games    Depth       NPS            Nodes
-------------------------------------------------------------------------------------
Lc0 Alpha-Beta      :      50     22            56750      5506278
Lc0 610383          :      50     25           175290     17496804
Stockfish 051121    :      50     46        111391469   4853958733
Scorpio v3.0.14d    :      50     12           162376      7141566
Ceres 0.95-rc4      :      50     12           176446     22319608
Dragon 2.5          :      50     41         87121728   4462082947

   # PLAYER              :  RATING  ERROR  POINTS  PLAYED   (%)  CFS(%)    W    D    L  D(%)
   1 Stockfish 051121    :       0   ----    28.5      50  57.0      65   14   29    7  58.0
   2 Lc0 610383          :     -14     71    27.5      50  55.0      64   10   35    5  70.0
   3 Ceres 0.95-rc4      :     -28     69    26.5      50  53.0      58    9   35    6  70.0
   4 Dragon 2.5          :     -34     67    26.0      50  52.0      94   11   30    9  60.0
   5 Lc0 Alpha-Beta      :     -89     72    22.0      50  44.0      84    7   30   13  60.0
   6 Scorpio v3.0.14d    :    -124     67    19.5      50  39.0     ---    3   33   14  66.0

White advantage = 136.78 +/- 16.42
Draw rate (equal opponents) = 100.00 % +/- 0.77

[Владимир001]

Deep Fritz 14 for learning ability for books
i7 980x over clocked
4 min each side T/C
6 cores per engine ponder off
1024 MB Hash
6 man syzygy on SSD
Late draw never selected to avoid short draws.
Windows 7 Pro
No large pages applied
All AV turned off
Engine Sugar 2.40 exp disabled NNUE+ Classical Enabled

32 books entered
Round one will consist of 1984 games.
Top ten books will continue to round 2 T/C and total games to decide at the time.
PGN of games will be given at the end of the tournament final rounds.

After 9 rounds tournament finished.

1- Predator 2 76,5 points
2- DON CTG 76,0 points
3- Annabelle2021 75.5 points


9000 games

https://www.mediafire.com/file/vzqge.../tour.rar/file

[Владимир001]

PGN - http://kirill-kryukov.com/chess/disc...e.php?id=49438

Развернуть для просмотра

CCRL 40/15 Rating List - Custom engine selection
1296384 games played by 2950 programs, run by 25 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 November 12, 2021 with Bayeselo based on 1'296'384 games
Tested by CCRL team, 2005-2021, http://ccrl.chessdom.com/ccrl/4040/

Rank                Engine                  Elo  +    -  Score  AvOp  Games
1 Clover 2.4 64-bit 4CPU                  3279  +27  -26  52.3%  -13.5  390
  Clover 2.3.1 64-bit 4CPU                3176  +26  -26  45.5%  +26.0  443

[свернуть]

[Владимир001]

PGN - http://kirill-kryukov.com/chess/disc...e.php?id=49439

Развернуть для просмотра

CCRL 40/15 Rating List - Custom engine selection
1296384 games played by 2950 programs, run by 25 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 November 12, 2021 with Bayeselo based on 1'296'384 games
Tested by CCRL team, 2005-2021, http://ccrl.chessdom.com/ccrl/4040/

Rank                Engine                  Elo  +    -  Score  AvOp  Games
1 Chiron 5 64-bit 4CPU                    3255  +27  -28  45.5%  +27.5  356
  Chiron 4 64-bit 4CPU                    3186  +11  -11  42.4%  +51.0  2540
  Chiron 3.01 64-bit 4CPU                3128  +24  -24  47.9%  +11.7  467
  Chiron 2 64-bit 4CPU                    3083  +15  -15  42.6%  +48.4  1485
  Chiron 1.5 64-bit 4CPU                  3045  +19  -19  42.2%  +46.7  856
  Chiron 1.1a 64-bit 4CPU                3039  +20  -20  42.6%  +46.1  789

[свернуть]

[Владимир001]

PGN - http://kirill-kryukov.com/chess/disc...e.php?id=49440

Развернуть для просмотра

CCRL 40/15 Rating List - Custom engine selection
1296384 games played by 2950 programs, run by 25 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 November 12, 2021 with Bayeselo based on 1'296'384 games
Tested by CCRL team, 2005-2021, http://ccrl.chessdom.com/ccrl/4040/

Rank                Engine                  Elo  +    -  Score  AvOp  Games
1 Chiron 5 64-bit                        3183  +24  -24  51.8%  -11.0  491
  Chiron 4 64-bit                        3114  +8  -8  45.1%  +33.3  5005
  Chiron 3.01 64-bit                      3050  +20  -20  45.6%  +27.1  776
  Chiron 2 64-bit                        3014  +13  -13  45.6%  +27.4  1923
  Chiron 1.1a 64-bit                      2977  +13  -13  45.5%  +26.6  1819
  Chiron 1.5 64-bit                      2963  +15  -15  45.4%  +27.8  1453
  Chiron 1.5 32-bit                      2939  +16  -16  46.0%  +23.1  1296
  Chiron 1.1a 32-bit                      2930  +14  -15  45.1%  +28.4  1558

[свернуть]