Разные матчи и турниры движков

[Владимир001]

Excalibur 1.0 wins 4 League JCER edition 41, 2020.12.19 - 2020.12.20

Temp 3' + 3" 180 games GUI-Fritz 17
Promotion to the 3 League: Excalibur 1.0, Mars 3.5.
Relegation to 5 League: ChessBrainVB 3.74, Senpai 2.0.

[Владимир001]

JCER CUP 2020 - Raubfisch and Zeus won ex aequo!

[Владимир001]

SPELL OF THE TWO LADIES

Intel i5 Quad
ChessGUI
256mb hash each
3-4-5 piece tablebases
Ponder off
GBSelect2019.cgb book
40 moves in 25 minutes repeating (adapted for the CCRL)
4 cycles 44 rounds
All engines 64-bit where available


Final Standings

27.5 - Googleplex Starthinker 1.6 64-bit
26.0 - Isa 2.0.83 64-bit
25.0 - Hedgehog 2.0 64-bit
24.5 - NoC 4.20 64-bit
23.0 - Ares GB 1.1 64-bit
22.5 - Myrddin 0.87 64-bit
22.0 - Jazz 840 64-bit
20.5 - CT800 1.41 64-bit
20.0 - RomiChess P3n 64-bit
19.5 - Ceibo 0.5 64-bit
17.5 - Zevra 2.1.2 r248 64-bit
16.0 - Barbarossa 0.6.0 64-bit


The complete tournament pgn (zipped) can be downloaded here:
http://kirill-kryukov.com/chess/disc...e.php?id=47572

[Владимир001]

Match: Match of Lc0 66719 vs Cfish NNUE 4CPU (Pure eval)

Развернуть для просмотра

Lc0-version: v0.26.3
Lc0-options : -Backend=cudnn -NNCacheSize=3000000 -MaxOutOfOrderEvalsFactor=2.4 -MaxCollisionEvents=256 -TwoFoldDraws=true -MinibatchSize=192
MLH-settings S19 SuFi: -MovesLeftMaxEffect=0.2 -MovesLeftSlope=0.004 -MovesLeftThreshold=0.0 -MovesQuadraticFactor=0.0 -MovesScaledFactor=1.0
Cfish-version: Cfish NNUE 301120 x64 bmi2 (30.11.2020), compile by Chessman
Cfish-options: threads=4, hash=4096 MB, NNUE=true, eval=Pure, net=nn-62ef826d1a6d.nnue
Time control: 10min + 5s
Hardware: notebook with mobile i7-7700HQ 4x2.8GHz (with HT), mobile Geforce GTX 1050 TI 4GB
Book: Balsa_Special.pgn, 10 plies, sequentially, colors reversed
Tablebases: Syzygy 5-men on SSD
Speed: around 25k nodes per move for J92/J94 // J92/J94: 1.0k nps, Cfish NNUE Pure 4CPU: 4975k nps
Adjudication: Draw at 50 moves rule, 3-fold repetition, after 200 moves, at 40 moves with eval = 0.00 ; 5-men Gaviota tablebase adjudication
Software: Arena 3.5.1, ponder=off
Elo rating: Ordo 1.2.6
Comments: 66719 added which has a result equal with J92-280. ERET test results included in the table.

   # PLAYER                      :  RATING  ERROR  POINTS  PLAYED   (%)  CFS(%)    W    D    L  D(%)  ERET test
   1 Cfish-NNUE 301120_Pure_4CPU :     0.0   ----   221.0     400  55.3     100   46  350    4  87.5
   2 Lc0.net.JH.J92-280          :   -31.7   22.2    45.5     100  45.5      50    3   85   12  85.0  80/111
   3 Lc0.net.66719               :   -31.7   22.6    45.5     100  45.5      66    0   91    9  91.0  83/111    <== NEW
   4 Lc0.net.JH.J94-80           :   -38.8   24.3    44.5     100  44.5      65    1   87   12  87.0  80/111
   5 Lc0.net.66613               :   -46.0   25.5    43.5     100  43.5     ---    0   87   13  87.0  81/111

[свернуть]

[Владимир001]

Match: Gauntlet SF201028 vs 66750

Развернуть для просмотра

Hardware: Geforce RTX 2070S & AMD Threadripper 3970X
Lc0 version: v0.26.3
Lc0 options: Backend=cuda-fp16 NNCacheSize=4000000 MinibatchSize=80 option.MaxOutOfOrderEvalsFactor=2.4 MovesLeftMaxEffect=0.2 MovesLeftThreshold=0.0 MovesLeftSlope=0.004 MovesLeftQuadraticFactor=0.0 MovesLeftScaledFactor=1.0 MovesLeftConstantFactor=0.0
SF options: Threads=8 Hash=4096
Speed: Lc0: 5.9 kN/s, SF201028: 13,275 kN/s (from benchmarks)
Time control: 10+0.1s
Book: openings-dd-8ply-2000.pgn, sequential, color reversed
Tablebases: 6-man syzygy
Adjudication: 6-man TB, -draw movenumber=35 movecount=5 score=15 -resign movecount=5 score=1000
Software: cutechess-cli
Comments: I decided to revert back to STC tests as the 5m+5s TC was too SSS. As Lc0 nps is lower with STC (whereas SF is largely unaffected), I have recalculated the ratio based on the much shorter time per move which gives 8 threads for SF rather than the previously used 23 threads.

   # PLAYER              :  RATING  ERROR  POINTS  PLAYED   (%)  CFS(%)     W     D     L
-  1 stockfish_201028    :     0.0   ----  7030.0   12000  58.6     100  2943  8174   883
   2 lc0.net.J94-80      :   -57.8    6.0  1680.5    4000  42.0      89   300  2761   939
+  3 lc0.net.66750       :   -63.1    5.9  1652.0    4000  41.3      73   306  2692  1002
   4 lc0.net.J92-310     :   -65.8    5.9  1637.5    4000  40.9     ---   277  2721  1002

White advantage = 54.35 +/- 1.77
Draw rate (equal opponents) = 74.64 % +/- 0.45

https://pixeldrain.com/u/AQUJuto8

[свернуть]

[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, 66736
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      :    14.2    3.0  10394.5   20000    52      53   4200  12389   3411    62
+  3 lc0.net.66736      :    14.0    2.9  10390.5   20000    52      98   4281  12219   3500    61
   4 lc0.net.66637      :     9.4    2.9  10263.0   20000    51      61   4083  12360   3557    62
   5 lc0.net.66512      :     8.9    2.9  10247.0   20000    51     100   3935  12624   3441    63
-  6 Stockfish.dev      :     0.0   ----  48293.0  100000    48     ---  17292  62002  20706    62

White advantage = 53.07 +/- 0.67
Draw rate (equal opponents) = 64.09 % +/- 0.16

PGN:
https://pixeldrain.com/u/J1bWaEcS

[свернуть]

[Владимир001]

SuFi_Test (games 600/1000)

   # PLAYER              :  RATING  ERROR  POINTS  PLAYED   (%)  CFS(%)    W    D    L  D(%)
   1 Lc0-j94-80 ZZ       :       4     10   151.5     300  50.5      50   21  261   18  87.0
   2 Lc0 T60-66740 ZZ    :       4     10   151.5     300  50.5      70   18  267   15  89.0
   3 Stockfish 221220    :       0      6   297.0     600  49.5     ---   33  528   39  88.0

White advantage = 31.63 +/- 4.88
Draw rate (equal opponents) = 91.76 % +/- 1.43
MEDIANS
Name                    Games     Depth       NPS           Nodes Per Move
-------------------------------------------------------------------------------------
Stockfish 221220    :      600     33        34448213   48736510
Lc0-j94-80 ZZ       :      300     12           33104     107622
Lc0 T60-66740 ZZ    :      300     12           32801     108435

https://pixeldrain.com/u/drE9xB4g

[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: 66740 added and ends up close to nearby net.
Games: https://gofile.io/d/FsAYLz

 # PLAYER            : RATING  ERROR  PLAYED  (%)   CFS    W      D     L    D(%)
 1 Stockfish 12 4CPU :   0.0   ----   18000  54.48  100  3187  13240  1573  73.56
 2 lc0-263-j94-80    : -26.5    7.8    2000  46.23   54   168   1513   319  75.65
 3 lc0-263-66740     : -27.1    8.0    2000  46.15   51   185   1476   339  73.80
 4 lc0-263-66727     : -27.2    8.4    2000  46.13   70   196   1453   351  72.65
 5 lc0-263-j96-28    : -30.1    7.2    2000  45.73   58   178   1473   349  73.65
 6 lc0-263-66555     : -31.1    8.1    2000  45.58   51   180   1463   357  73.15
 7 lc0-263-j92-330   : -31.3    8.1    2000  45.55   53   183   1456   361  72.80
 8 lc0-263-j96-44    : -31.6    7.4    2000  45.50   62   163   1494   343  74.70
 9 lc0-263-66594     : -33.4    8.8    2000  45.25   98   173   1464   363  73.20
10 lc0-263-J92-190   : -45.5    7.0    2000  43.55  ---   147   1448   405  72.40

White advantage = 7.19 +/- 1.39
Draw rate (equal opponents) = 74.82 % +/- 0.30

[свернуть]

[Владимир001]

3 min 2 core frıtz16 gui syzgyz 3-4-5 nnue i7 3820 4ghz play 120 games

vvvvvvvvvvvv, Blitz 3min  2020

                                   1        2        3        4        5        6        
1   Stockfish 231220               ******** ½½½½½½½½ ½½½½½½½½ ½1½½½½½½ ½½½½½½½½ ½½1½½½½½  21.0/40
2   Cfish 131220 64 POPCNT NUMA    ½½½½½½½½ ******** ½½½½½½½½ ½½½½½½½½ ½½½½½½1½ ½½½½½½½½  20.5/40
3   Zeus 11.0                      ½½½½½½½½ ½½½½½½½½ ******** ½½½½½½½½ ½½½½½½½½ ½½½½½½½½  20.0/40  400.00
4   AI 15.0                        ½0½½½½½½ ½½½½½½½½ ½½½½½½½½ ******** ½½½½½½½½ ½½½1½½½½  20.0/40  399.00
5   Eman 6.61 64-bit SSE41,POPCNT  ½½½½½½½½ ½½½½½½0½ ½½½½½½½½ ½½½½½½½½ ******** ½½½½½½½½  19.5/40
6   Cfish 221220 64 POPCNT         ½½0½½½½½ ½½½½½½½½ ½½½½½½½½ ½½½0½½½½ ½½½½½½½½ ********  19.0/40

games
https://pixeldrain.com/u/hqJcas1b

[Владимир001]

Topple 0.8.0 wins JCER - Test new engines, 2020.12.23 - 2020.12.24

Temp 3' + 3" 132 games

[Владимир001]

BBC 1.4 SF NNUE wins JCER League E5 edition 41, 2020.12.22 - 2020.12.24

Temp 3' + 3" 306 games GUI-Fritz 15
Promotion to Group E4: BBC 1.4 SF NNUE, Supernova 2.2.1.
Relegation to Group E6: Jinx 1.0, Baislicka dev ac327fd, Frenzee 3.0.

[Владимир001]

V3.0.0 (219) тест

Vafra 3.0.0 против LeelaZero v0.26.2; игра в шахматы с медленным контролем времени на ноутбуке HP Pavilion dv7, использующем Intel® Core ™ i7, с четырьмя гиперпоточными ядрами, без использования графического процессора, поскольку графические процессоры AMD Accelerated Parallel Processing Caicos работали хуже, чем CPU; оба двигателя думали только о своем ходу, то есть 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; и Вафра, и Лила имели доступ к финальным столам, Вафра - ко всем пятеркам и некоторым шестерням, Лила - ко всем пятеркам; Лила использовала мозг 15 декабря 2020 г. NN66647; 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...Leela_342.html

[Владимир001]

Match: Lc0 26.3 vs Stockfish 181220
Nodes per move: Lc0 = 2,000 nodes. Stockfish = 1,000,000 nodes.
Hardware: RTX 3080, i9-10900kf
Opening book: Polh's Unbalanced Human Openings 6-moves (3314 pos.)

Развернуть для просмотра

Stockfish bmi2: threads=4, hash=256
Lc0 cuda-fp16: threads=1, smart-pruning-factor=1, mlh=on
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
Comments: 66792 did beat 66740 by half a point(!) as the new net ends up in between J92-330 and J94-80. Otherwise, the opening set generates a very low draw rate.
Games: https://gofile.io/d/WU4IiZ

# PLAYER               :  RATING ERROR  PLAYED  (%)   CFS    W     D      L    D(%)
1 sf-181220-4T-1000k   :    0.0  ----   26512  52.04  100  7758  12079  6675  45.56
2 lc0-263-j94-80-2k    :  -15.3   5.8    6628  48.25   71  1677   3042  1909  45.90
3 lc0-263-66792-2k     :  -17.7   6.2    6628  47.98   51  1681   2998  1949  45.23
4 lc0-263-66740-2k     :  -17.8   5.0    6628  47.97   79  1691   2977  1960  44.92
5 lc0-263-j92-330-2k   :  -20.8   5.9    6628  47.63  ---  1626   3062  1940  46.20

White advantage = 168.41 +/- 1.72
Draw rate (equal opponents) = 59.48 % +/- 0.42

[свернуть]

[Maratka]

Ну что сказать, фиг его знает, что это за сети (тут нужен некий контрольный турнир между ними), но 4 ядра Core против RTX 3080 - это явно форма за LZ, насколько я себе представляю скорость тензорных ядер второго поколения.

Т.е. если сети реально околотоповые, то в финале TCEC LZ будет несладко, если дойдет (т.е. DbK не сможет занять второе место в отборе).

Как-то так я это вижу...

p.s.
Nodes per move: Lc0 = 2,000 nodes. Stockfish = 1,000,000 nodes.

Вот тут бы понять, как оно соотносится...
Т.е. нужен матч скажем тех же LZ с теми же настройками, но против SF12, тогда динамика будет видна.

[Vizvezdenec]

Цитата: Сообщение от Maratka

Ну что сказать, фиг его знает, что это за сети (тут нужен некий контрольный турнир между ними), но 4 ядра Core против RTX 3080 - это явно форма за LZ, насколько я себе представляю скорость тензорных ядер второго поколения.

Т.е. если сети реально околотоповые, то в финале TCEC LZ будет несладко, если дойдет (т.е. DbK не сможет занять второе место в отборе).

Как-то так я это вижу...

p.s.
Nodes per move: Lc0 = 2,000 nodes. Stockfish = 1,000,000 nodes.

Вот тут бы понять, как оно соотносится...
Т.е. нужен матч скажем тех же LZ с теми же настройками, но против SF12, тогда динамика будет видна.

Стоп стоп стоп, там число узлов фиксированное, 2000 для LZ и 1000000 для SF (на ход).
Что ещё большая чепуха и я, кстати, совершенно не понимаю, зачем так делать - вырубать тайм менеджмент, да и в принципе...

[Владимир001]

Time control: 60 seconds / game + 0.6 seconds / move


Playing conditions

CPU: Dual Intel Xeon E5-2670v3 @ 2.6 GHz, 24 Cores
GPU: NVidia GeForce RTX 2070
OS: Windows 10 64-Bit
Tool: Cutechess-Cli
Leela Ratio: ~ 1.0
A/B-Engines: 16 Cores, 64 Bit PEXT/BMI2, default settings
NN-Engines: default settings
Hash-Table: A/B-Engines 1024 MB
NN-Cache: default settings
Lc0-backend: cudnn-fp16
Time control: 60 seconds + 0.6 seconds
Tablebases: No, but tablebase adjudication with 6 pieces
Openings: Hert_250_lowdraws.epd, changing colors
Ponder: Off
Learning: Off
Large Memory Pages: Off


December 21, 2020

Ordo v1.2.6: Fire 7.1 3200

New: 10. Lc0 0.26.3 66625

http://www.fastgm.de/16-60-0.6.html

[Владимир001]

4min+2sec, ponder ON

Testing Cfish vs Lc0 ; ponder =on, Parameters Lc0 v0.26.3
TCEC SuFi 19 MLH settings
Hardware : ,--backend=cuda-fp16
CPU : Intel E5-2697A v4 (1 x 16 cores, 3,1 GHz), 64 GB RAM,--threads=2
GPU : 1 x RTX 2060 (1800MHz),--nncache=20000000
--move-overhead=15000
https://docs.google.com/spreadsheets...gid=1294765898

[Владимир001]

MATCH: J94 & T66740 v SF Bullet
Hardware: i9-10980XE CPU 36 threads, 64GB, gpu RTX 3080 + 2070
LC0-version: v0.26.3
LC0-options: ZZ Tune 2
Software: Octagon
Time control: 1 minutes + 1 seconds
Book: Chad 6 ply
Tablebases: 3/4/5/6 piece

Развернуть для просмотра

Adjudication: -draw movenumber=35 movecount=10 score<=0.15 -resign movecount=10 score >=10
6 piece tablebase adjudication
Comment 1: Lc0 JH94-80 and T66740 equal result!
Comment 2: Pretty amazing result at bullet tc

   # PLAYER              :  RATING  ERROR  POINTS  PLAYED   (%)  CFS(%)    W    D    L  D(%)
   1 SF 221220           :       0      5   501.0    1000  50.1      55   54  895   52  89.5
   2 Lc0-j94-80 ZZ       :      -1      7   249.5     500  49.9     ---   26  447   27  89.5
   3 Lc0 T60-66740 ZZ    :      -1      7   249.5     500  49.9     ---   26  447   27  89.5

White advantage = 0.00 +/- 0.00
Draw rate (equal opponents) = 89.50 % +/- 0.97
MEDIANS
Name                    Games     Depth       NPS           Nodes
-------------------------------------------------------------------------------------
Stockfish 221220    :     1000     34        34556872   48797655
Lc0-j94-80 ZZ       :      500     12           33542     108292
Lc0 T60-66740 ZZ    :      500     12           32968     109954

https://pixeldrain.com/u/fGXvwbKj

[свернуть]

[Владимир001]

https://sites.google.com/site/comput.../candidate2020

Партии можно скачать по этой ссылке

[Владимир001]

http://www.jurjevic.org.uk/chess/gam...Leela_343.html

[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: J94-80, J98.3-8, 66512, 66592, 66637, 66736
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.J94-80     :    16.3    3.0  10455.0   20000    52      77   4253  12404   3343    62
   2 lc0.net.J98.3-8    :    14.8    3.0  10412.0   20000    52      61   4207  12410   3383    62
   3 lc0.net.66592      :    14.1    3.0  10394.5   20000    52      53   4200  12389   3411    62
   4 lc0.net.66736      :    14.0    3.0  10390.5   20000    52      98   4281  12219   3500    61
   5 lc0.net.66637      :     9.4    2.9  10263.0   20000    51      61   4083  12360   3557    62
   6 lc0.net.66512      :     8.9    3.0  10247.0   20000    51     100   3935  12624   3441    63
-  7 Stockfish.dev      :     0.0   ----  57838.0  120000    48     ---  20635  74406  24959    62

White advantage = 52.76 +/- 0.60
Draw rate (equal opponents) = 64.08 % +/- 0.15

PGN:
https://pixeldrain.com/u/1M3vE1zG

[свернуть]

[Владимир001]

Этот тест https://sites.google.com/site/comput...t-4men-vs-5men

показывает, как хеширование влияет на производительность движка. Чем больше хеш, чем он должен быть, тем хуже будет производительность движка.

[Maratka]

Только это почему-то показано для 6 ядер и трех минут на партию.
Как насчет вполне себе типичного для сегодня 8-12 ядер с двумя потоками на ядро, и 60 минут?

[Владимир001]

SugaR AI 1.20 wins JCER Test Tournament, 2020.12.22 - 2020.12.26

Temp 3' + 3" 306 games GUI-Fritz 17

[Владимир001]

https://www.chess2u.com/t15482-ether...match-with-ltc

[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: 66800 takes the lead! LS-15 added.
Games: https://gofile.io/d/JB5sl9

# PLAYER              :  RATING ERROR  PLAYED  (%)   CFS    W     D     L   D(%)
1 lc0-263-66800       :   14.5   13.1   1000  51.95   67   215   609   176  60.90
2 lc0-263-j94-80      :   10.4   12.2   1000  51.40   56   214   600   186  60.00
3 lc0-263-66740       :    9.0   13.7   1000  51.20   76   221   582   197  58.20
4 lc0-263-66662       :    2.6   12.2   1000  50.35   66   209   589   202  58.90
5 stockfish-181220-8T :    0.0   ----   6000  49.96   66  1218  3559  1223  59.32
6 lc0-263-j92-280     :   -2.6   12.6   1000  49.65  100   196   601   203  60.10
7 lc0-263-ls-15       :  -32.1   11.8   1000  45.70  ---   168   578   254  57.80

White advantage = 87.63 +/- 2.97
Draw rate (equal opponents) = 64.76 % +/- 0.66

[свернуть]

[Владимир001]

http://www.jurjevic.org.uk/chess/gam...Leela_345.html

[Владимир001]

Sctr 1.1f wins 5 League JCER edition 41, 2020.12.23 - 2020.12.26

Temp 3' + 3" 180 games, GUI-Fritz 17
Promotion to the 4 League: Sctr 1.1f , SmarThink 1.98.
Relegation to 6 League: FabChess 1.16, Counter 3.5.

[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: 66683 now on top! 66680 ends up logically inbetween 66662 and 66740. Up to now, T60 results follow net ID suggesting progress through time, but CFS is low.
Games: https://gofile.io/d/8RDO5B

# PLAYER              : RATING ERROR  PLAYED  (%)   CFS    W     D     L   D(%)
1 lc0-263-66833       :  16.4   14.3   1000  52.20   58   221   602   177  60.20
2 lc0-263-66800       :  14.5   11.2   1000  51.95   69   215   609   176  60.90
3 lc0-263-j94-80      :  10.4   13.2   1000  51.40   56   214   600   186  60.00
4 lc0-263-66740       :   8.9   13.5   1000  51.20   52   221   582   197  58.20
5 lc0-263-66680       :   8.6   12.4   1000  51.15   78   210   603   187  60.30
6 lc0-263-66662       :   2.6   11.0   1000  50.35   68   209   589   202  58.90
7 stockfish-181220-8T :   0.0   ----   8000  49.55   64  1582  4764  1654  59.55
8 lc0-263-j92-280     :  -2.6   14.5   1000  49.65  100   196   601   203  60.10
9 lc0-263-ls-15       : -32.1   13.2   1000  45.70  ---   168   578   254  57.80

White advantage = 86.87 +/- 2.73
Draw rate (equal opponents) = 64.93 % +/- 0.59

[свернуть]

[Владимир001]

Monolith 2 wins 6 League JCER edition 41, 2020.12.26 - 2020.12.27

Temp 3' + 3" 180 games GUI-Fritz 17
Promotion to the 5 League: Monolith 2, Ippos 1.2.7.
Relegation to Group E1: Gaviota 1.0, Francesca 0.30.