Разные матчи и турниры движков

[Владимир001]

https://www.chess2u.com/t16758-fatfr...cpu-ltc#103481

[Владимир001]

Gauntlet: SF-nnue vs PR1527 & master

Развернуть для просмотра

LC0 version (master): b1bf3f3
LC0 version (pr1527): 3411e4b
LC0 options (master): threads=1, minibatch-size=192, time-manager=legacy
LC0 options (pr1527): threads=1, minibatch-size=192, time-manager=smooth
SF options: defaults
LC0 Time control: 1s+0.1s
SF Time control: 1s+0.025s
LC0 Net: 703810
SF Net: nn-62ef826d1a6d
Book: custom-3k (1-3 ply)
Hardware: gtx970 + i5-4570
Tablebases: syzygy 5-man
Adjudication: -draw movenumber=50 movecount=5 score=10 -resign movecount=5 score=700 twosided=true
Software: cutechess-cli
Comment: smooth tm did better, but had a few losses on time. didnt see any time losses with legacy tm @crem

[свернуть]

   # PLAYER                   :  RATING  ERROR  POINTS  PLAYED   (%)  CFS(%)    W     D    L  D(%)
   1 lc0-pr1527 (smooth)      :     8.4   14.4   640.0    1250    51      87  339   602  309    48
   2 sf                       :     0.0   ----  1245.0    2500    50      79  653  1184  663    47
   3 lc0-master (legacy)      :    -5.6   13.8   615.0    1250    49     ---  324   582  344    47

White advantage = -4.21 +/- 5.11
Draw rate (equal opponents) = 47.39 % +/- 1.04

[Владимир001]

Match Ceres v0.88 vs LC0 0.27 RC1 (with multigather enabled)
Network j94-100

Развернуть для просмотра

LC0 options: --nodes-as-playouts --multi-gather --move-overhead=250 --max-out-of-order-evals-factor=2.4 --max-collision-events=500 --max-collision-visits=500 -w d:\weights\lczero.org\j94-100.pb.gz -t 2 --nncache=5000000 --backend=cuda-fp16 --backend-opts=gpu=0 --syzygy-paths="f:\sygyzy\5and6man"
Hardware: Intel 6142 Processor, each game on one GPU from pool of 2 A100, 1 Titan RTX, 1 Titan X
Time control: 180 seconds plus 1 second increment
Software: Integrated Ceres tournament manager
Opening book: TCEC18 + TCEC19
Tablebases: Syzygy 5 and 6-men
Comment: This test is a identical to yesterday's test, except with 3 non-default parameters provided on command line to optimize multigather for this configutation, (based on advice from Tilps).
Comment: With these optimizations, LC0 nearly equalizes in NPS and more than equalizes in Elo.
Comment: However two observations:
Comment: - Ceres achieves its results across all configurations (hardware, network, time control) "out of the box" by using heuristics and auto-tuning to avoid the need for user knowledge or experimentation with parameters to achieve good results. Most likely LC0 could improve in this area without much additional effort.
Comment: - These tests were run in a configuration yielding a fairly modest 17K nps. With smaller networks or multiple high-end GPUs, Ceres regains some NPS advantage (e.g. at 50K raw GPU nps the search nps is higher by about 25%). Perhaps the "dual overlapped executors" approach of Ceres could help LC0.

[свернуть]

   # PLAYER    :  RATING  ERROR  POINTS  PLAYED   (%)  CFS(%)
   1 LC0_0     :      15     12   106.0     204  52.0      88
   2 Ceres1    :       0     12    98.0     204  48.0     ---

White advantage = 98.92 +/- 12.55
Draw rate (equal opponents) = 82.08 % +/- 4.76

[Владимир001]

https://www.chess2u.com/t16758-fatfr...cpu-ltc#103494

[Владимир001]

Fisherov 0.95 wins JCER - Lucas HP Test Tournament, 2021.02.12-14

Temp 3' + 3" GUI-Lucas Chess

[Владимир001]

https://www.chess2u.com/t16758-fatfr...cpu-ltc#103512

[Владимир001]

https://www.chess2u.com/t16758-fatfr...cpu-ltc#103517

[Владимир001]

Valentines Championship (First Round DRR)

   # PLAYER                  :  RATING  ERROR  POINTS  PLAYED   (%)  CFS(%)    W    D    L  D(%)
+1 Fat Fritz 2             :      93    188     6.5      10  65.0      67    5    3    2  30.0
 2 v0.27.0-rc1_J94-100     :      31    183     5.5      10  55.0      50    5    1    4  10.0
 3 Stockfish 140221        :      31    185     5.5      10  55.0      58    4    3    3  30.0
 4 Dragon                  :       0    187     5.0      10  50.0      58    4    2    4  20.0
 5  v0.27.0-rc1_T67673     :     -30    180     4.5      10  45.0      74    3    3    4  30.0
 6  v0.27.0-rc1_SV-4619    :    -125    198     3.0      10  30.0     ---    1    4    5  40.0

White advantage = 0.00 +/- 0.00
Draw rate (equal opponents) = 28.26 % +/- 9.86

https://www.twitch.tv/navratil25

[Владимир001]

Honey 12-R1 wins JCER - Lucas GHI Tournament, 12-14.02.2021

Temp 3' + 3" GUI-Lucas Chess

[Владимир001]

https://www.chess2u.com/t16758-fatfr...cpu-ltc#103545

[Владимир001]

Hardware: i5-8400 CPU @ 2.80GHz, 6 cores, 32GB, gpu RTX 2060, 6GB
Gauntlet: SF-dev vs. LC0 384x30b nets
Network IDs: J94-80, J94-100, J94-130, J98.3-8, J98.1-16, 66512, 66592, 66637, 66736, 66740, 66887, 66937, 67145, 67192, 67225, 67336, 67383, 67436, 67483, 67528, 67577, 67616, 67646, 67670
LC0-version: v0.26.3

Развернуть для просмотра

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: 20K more games for J94-100 with adjudication rule -draw movenumber=35 movecount=5 score=8 -resign movecount=5 score=650
67670 is very low. Sampling fluctuations?

[свернуть]

   # PLAYER              :  RATING  ERROR    POINTS  PLAYED   (%)  CFS(%)      W       D       L  D(%)
-  1 lc0.net.J94-100     :    21.1    2.1   21176.0   40000    53      75   8684   24984    6332    62
   2 lc0.net.67646       :    19.8    2.9   10553.0   20000    53      59   4381   12344    3275    62
   3 lc0.net.67436       :    19.4    2.9   10540.0   20000    53      52   4377   12326    3297    62
   4 lc0.net.67528       :    19.3    2.9   10537.0   20000    53      56   4331   12412    3257    62
   5 lc0.net.J94-130     :    18.9    3.0   10528.0   20000    53      78   4314   12428    3258    62
   6 lc0.net.67616       :    17.3    2.9   10481.5   20000    52      52   4298   12367    3335    62
   7 lc0.net.67336       :    17.2    3.0   10478.5   20000    52      53   4293   12371    3336    62
   8 lc0.net.67483       :    17.0    2.9   10474.0   20000    52      58   4336   12276    3388    61
   9 lc0.net.67145       :    16.6    3.0   10462.5   20000    52      52   4265   12395    3340    62
  10 lc0.net.67192       :    16.5    3.0   10459.0   20000    52      53   4320   12278    3402    61
- 11 lc0.net.J94-80      :    16.3    3.0   10455.0   20000    52      50   4253   12404    3343    62
  12 lc0.net.67383       :    16.3    2.9   10454.5   20000    52      52   4292   12325    3383    62
+ 13 lc0.net.67670       :    16.2    3.0   10451.0   20000    52      53   4266   12370    3364    62
  14 lc0.net.67577       :    16.0    3.0   10446.5   20000    52      61   4252   12389    3359    62
  15 lc0.net.J98.1-16    :    15.4    2.9   10430.0   20000    52      53   4181   12498    3321    62
  16 lc0.net.67225       :    15.2    3.0   10425.0   20000    52      59   4254   12342    3404    62
  17 lc0.net.J98.3-8     :    14.8    2.9   10412.0   20000    52      62   4207   12410    3383    62
  18 lc0.net.66592       :    14.2    3.0   10394.5   20000    52      53   4200   12389    3411    62
  19 lc0.net.66736       :    14.0    2.9   10390.5   20000    52      59   4281   12219    3500    61
- 20 lc0.net.66740       :    13.5    3.0   10377.0   20000    52      57   4223   12308    3469    62
  21 lc0.net.66937       :    13.1    3.1   10366.0   20000    52      59   4198   12336    3466    62
  22 lc0.net.66887       :    12.6    3.0   10352.0   20000    52      93   4189   12326    3485    62
  23 lc0.net.66637       :     9.4    3.0   10263.0   20000    51      61   4083   12360    3557    62
  24 lc0.net.66512       :     8.9    3.0   10247.0   20000    51     100   3935   12624    3441    63
- 25 Stockfish.dev       :     0.0   ----  238846.5  500000    48     ---  84106  309481  106413    62

White advantage = 53.03 +/- 0.30
Draw rate (equal opponents) = 64.04 % +/- 0.07

https://pixeldrain.com/u/NxpURpaS

[Владимир001]

CorChess 270121 wins JCER - Strong New Engines Test, 2021.02.14 - 2021.02.15

Temp 3' + 3" 168 games GUI-Fritz 17

[Владимир001]

Ice4 853 wins 4 League JCER edition 43, 2021.02.10 - 2021.02.12

Temp 3' + 3" 180 games GUI-Fritz 17
Promotion to the 3 League: Ice4 853, Equinox 3.30.
Relegation to 5 League: Senpai 2.0, Ippos 1.2.7.

[Владимир001]

https://www.chess2u.com/t16758-fatfr...cpu-ltc#103551

[Владимир001]

Valentines Championship Round 2 (60/300 games)

   # PLAYER                  :  RATING  ERROR  POINTS  PLAYED   (%)  CFS(%)    W    D    L  D(%)
   1 Fat Fritz 2             :      45     84    11.5      20  57.5      59    5   13    2  65.0
   2 Stockfish 140221        :      30     83    11.0      20  55.0      50    5   12    3  60.0
   3 v0.27.0-rc1_J94-100     :      30     84    11.0      20  55.0      75    6   10    4  50.0
   4 Dragon                  :     -15     82     9.5      20  47.5      59    4   11    5  55.0
   6 v0.27.0-rc1_SV-4619     :     -60     86     8.0      20  40.0     ---    1   14    5  70.0

White advantage = 0.00 +/- 0.00
Draw rate (equal opponents) = 62.12 % +/- 7.09

MEDIANS
Name                        Games    Depth       NPS           Nodes
-------------------------------------------------------------------------------------
v0.27.0-rc1_T67673      :      20     21           53436     3885078
Dragon                  :      20     32        35120355   880383052
Stockfish 140221        :      20     46        35016747   819020230
v0.27.0-rc1_SV-4619     :      20     18           66149     3076796
Fat Fritz 2             :      20     50        31344065   701906643
v0.27.0-rc1_J94-100     :      20     15           56819     2999246

https://www.twitch.tv/navratil25

[Владимир001]

Gauntlet: SF-nnue vs PR1528 & master
LC0 version (master): 9fa7bfa
LC0 version (pr1528): 18f61b5
LC0 options (master): threads=1, minibatch-size=192, time-manager=legacy
LC0 options (pr1528): threads=1, minibatch-size=192, time-manager=smooth
SF options: defaults
LC0 Time control: 1s+0.1s
SF Time control: 1s+0.025s
LC0 Net: 703810
SF Net: nn-62ef826d1a6d
Book: custom-3k (1-3 ply)
Hardware: gtx970 + i5-4570
Tablebases: syzygy 5-man
Adjudication: -draw movenumber=50 movecount=5 score=10 -resign movecount=5 score=700 twosided=true
Software: cutechess-cli
Comment:

   # PLAYER                :  RATING  ERROR  POINTS  PLAYED   (%)  CFS(%)     W     D     L  D(%)
   1 lc0-pr1528 (smooth)   :    18.2    8.6  1840.5    3500    53     100  1014  1653   833    47
   2 sf                    :     0.0   ----  3463.0    7000    49      99  1816  3294  1890    47
   3 lc0-master (legacy)   :   -10.7    8.3  1696.5    3500    48     ---   876  1641   983    47

White advantage = -10.94 +/- 3.11
Draw rate (equal opponents) = 47.18 % +/- 0.61

https://pixeldrain.com/u/qxtP35mo

[Viktor]

В моих тестах версия J94-120 хорошо себя проявляла. Даже лучше, чем J94-100. Интересно их стравить.

[Владимир001]

THE RETURN OF ICARUS

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 36 rounds
All engines 64-bit where available


Final Standings

22.5 - Waxman 2020
22.0 - Myrddin 0.87 64-bit
20.5 - Simplex 0.9.8 64-bit
20.0 - Ares GB 1.1 64-bit
19.0 - Zeus 1.29
17.0 - Barbarossa 0.6.0 64-bit
16.5 - KnightDreamer 3.3
15.0 - Isa 2.0.83 64-bit
14.0 - EveAnn 1.72
13.5 - CT800 1.41 64-bit


The complete tournament pgn (zipped) can be downloaded here:

http://kirill-kryukov.com/chess/disc...e.php?id=47890

[Владимир001]

https://www.chess2u.com/t16758-fatfr...cpu-ltc#103562

[Владимир001]

https://www.chess2u.com/t16758-fatfr...cpu-ltc#103570

[Владимир001]

# PLAYER                :  RATING  ERROR  POINTS  PLAYED   (%)  CFS(%)    W     D    L  D(%)
   1 lc0-pr1530 (smooth)   :    19.0    9.2  1581.0    3000    53     100  877  1408  715    47
   2 sf                    :     0.0   ----  1419.0    3000    47     ---  715  1408  877    47

[Владимир001]

Round Robin: lc0.net.720616 vs lc0.net.lc0.net.740552
LC0-version: lc0-v0.27-rc1, Backend=cuda-fp16, Threads=2, Default parameters
Hardware: RTX 2060
Software: Cutechess-CLI
Time control: 1000 nodes/move
Book: openings-12ply-100k.pgn, sequential, color reversed
Tablebases: Syzygy 5-men
Adjudication: -draw movenumber=50 movecount=5 score=8 -resign movecount=5 score=1000
Comment: lc0.net.720616 last-T72-LR 0.04

  # PLAYER            :  RATING  ERROR  POINTS  PLAYED     W    L     D  D(%)  CFS(%)
+ 1 lc0.net.740552    :    10.6   14.9   515.0    1000   252  222   526    53      92
- 2 lc0.net.720616    :     0.0   ----  2031.0    4000  1008  946  2046    51      55
  3 lc0.net.740430    :    -1.1   15.2   498.5    1000   240  243   517    52      74
  4 lc0.net.740329    :    -8.5   15.4   488.0    1000   227  251   522    52      91
  5 lc0.net.740273    :   -23.1   15.3   467.5    1000   227  292   481    48     ---

White advantage = 37.88 +/- 3.79
Draw rate (equal opponents) = 51.85 % +/- 0.79

[Владимир001]

Match: T74 vs t72 improvement
LC0-version: 0.26.3
LC0 options: Network Id: 740554 vs 720616
Options: Both: MinibatchSize=64, mlhtreshold = 0;
Hardware: 2threads, gtx1080
Time control: 800 nodes/move
Book: book-ply4-unifen-Q-0.0-0.25.pgn
Tablebases: 6-man syzygy
Adjudication: Draw adjudication: move number: 50, move count: 5, score: 10 cp, Resign adjudication: Move count: 5, score: 1000cp, Tablebase adjudication
Software: Cutechess-gui
ordo: ./ordo-win64.exe -Q -N 0 -D -a 0 -A "lc0 t74" -W -n4 -s500 -U "0,1,2,3,4,5,6,7,8,9,10" -j h2h.txt -C matrix.csv -p results.pgn
Comment: Compare between t74 and t72 last in same lr

   # PLAYER     :  RATING  ERROR  POINTS  PLAYED   (%)  CFS(%)    W    D    L  D(%)
   1 lc0 t74    :       0   ----  1028.5    2000  51.4      96  557  943  500  47.1
   2 lc0 t72    :     -10     11   971.5    2000  48.6     ---  500  943  557  47.1

White advantage = 25.48 +/- 5.91
Draw rate (equal opponents) = 47.42 % +/- 1.12

https://pixeldrain.com/u/yzGEHPaq

[Владимир001]

https://www.chess2u.com/t16758-fatfr...cpu-ltc#103592

[Владимир001]

https://www.chess2u.com/t16758-fatfr...cpu-ltc#103601

[Владимир001]

Round Robin: lc0.net.720616 vs lc0.net.lc0.net.740585
LC0-version: lc0-v0.27.0-rc2, Backend=cuda-fp16, Threads=2, Default parameters
Hardware: RTX 2060
Software: Cutechess-CLI
Time control: 1000 nodes/move
Book: openings-8ply-10k.pgn, sequential, color reversed
Tablebases: Syzygy 5-men
Adjudication: -draw movenumber=50 movecount=5 score=8 -resign movecount=5 score=1000
Comment: lc0.net.720616 last-T72-LR 0.04

    # PLAYER            :  RATING  ERROR  POINTS  PLAYED    W    L     D  D(%)  CFS(%)
 1 lc0.net.740585    :     7.2   11.0  1020.5    2000  519  478  1003    50      90
 2 lc0.net.720616    :     0.0   ----   979.5    2000  478  519  1003    50     ---

White advantage = 32.88 +/- 5.54
Draw rate (equal opponents) = 50.62 % +/- 1.14

https://pixeldrain.com/u/gABhrNM3

[Владимир001]

Valentines Championship Round 3 (90/300 games)

Leela is killing it!
I can't wait to try out v0.27.0-rc2 with Smooth time manager

 # PLAYER                  :  RATING  ERROR  POINTS  PLAYED   (%)  CFS(%)    W    D    L  D(%)
+1 v0.27.0-rc1_J94-100     :      49     64    17.5      30  58.3      50    9   17    4  56.7
 2 Fat Fritz 2             :      49     64    17.5      30  58.3      83    7   21    2  70.0
 3 Stockfish 140221        :       0     64    15.0      30  50.0      72    5   20    5  66.7
 4 v0.27.0-rc1_SV-4619     :     -30     64    13.5      30  45.0      50    3   21    6  70.0
 5 v0.27.0-rc1_T67673      :     -30     64    13.5      30  45.0      58    5   17    8  56.7
 6 Dragon                  :     -40     63    13.0      30  43.3     ---    4   18    8  60.0


MEDIANS
Name                        Games    Depth       NPS           Nodes
-------------------------------------------------------------------------------------
v0.27.0-rc1_T67673      :      31     18           54134     2992878
Dragon                  :      31     33        31864926   764381112
Stockfish 140221        :      30     46        34117770   750056366
v0.27.0-rc1_SV-4619     :      30     17           57744     2853220
Fat Fritz 2             :      30     49        30576800   651423047
v0.27.0-rc1_J94-100     :      30     15           57996     2758828

https://www.twitch.tv/navratil25

[Владимир001]

https://www.chess2u.com/t16758-fatfr...cpu-ltc#103614

[Владимир001]

https://www.chess2u.com/t16758-fatfr...cpu-ltc#103628

[Владимир001]

Fisherov 0.96 wins JCER Test - 2021.02.16 - 2021.02.18

Temp 3' + 3"