Stockfish 13 - 16 vs 32 threads

Discussion of computer chess matches and engine tournaments.

Moderators: hgm, Rebel, chrisw

fastgm
Posts: 818
Joined: Mon Aug 19, 2013 6:57 pm

Stockfish 13 - 16 vs 32 threads

Post by fastgm »

Playing conditions

Code: Select all

CPU:          Dual Intel Xeon E5-2673 v4, 40 cores (80 threads)
OS:           Windows 10 64-Bit
Tool:         Cutechess-Cli, no draw and resign rules
Engines:      Stockfish 13 x64 BMI2, default settings
Hash-Table:   2048 MB
Time control: 60 seconds + 0.6 seconds
Tablebases:   None
Openings:     4 moves book, changing colors
Ponder:       Off
Stockfish 13 x64 BMI2 32 threads vs Stockfish 13 x64 BMI2 16 threads, 1000 games

+20 Elo

Code: Select all

Stockfish 13 T32   +20  +67/=923/-10 52.85%  528.5/1000
Stockfish 13 T16   -20  +10/=923/-67 47.15%  471.5/1000

Engine                 Depth       Time   Games     Moves  Average    MIDG   EARLY    ENDG    LATE
Stockfish 13 T32       39.78   26:09:09    1000     76100    1.24    32.34 | 32.95 | 40.31 | 51.18
Stockfish 13 T16       37.19   26:12:55    1000     76061    1.24    30.22 | 30.57 | 37.05 | 48.62

Legend

Depth     : Overall average search depth
Time      : Total time engine used
Moves     : Total moves engine played
Average   : Average time per move in centi-seconds

MIDG      : Average search depth during the opening and middle game
EARLY     : Average search depth during the middle game without Queens, the so called early end game
ENDG      : Average search depth during the end game
LATE      : Average search depth during the late end game
Top
User avatar
pohl4711
Posts: 2432
Joined: Sat Sep 03, 2011 7:25 am
Location: Berlin, Germany
Full name: Stefan Pohl

Re: Stockfish 13 - 16 vs 32 threads

Post by pohl4711 »

fastgm wrote: ↑Tue Apr 06, 2021 8:52 pm Playing conditions

Code: Select all

CPU:          Dual Intel Xeon E5-2673 v4, 40 cores (80 threads)
OS:           Windows 10 64-Bit
Tool:         Cutechess-Cli, no draw and resign rules
Engines:      Stockfish 13 x64 BMI2, default settings
Hash-Table:   2048 MB
Time control: 60 seconds + 0.6 seconds
Tablebases:   None
Openings:     4 moves book, changing colors
Ponder:       Off
Stockfish 13 x64 BMI2 32 threads vs Stockfish 13 x64 BMI2 16 threads, 1000 games

+20 Elo

Code: Select all

Stockfish 13 T32   +20  +67/=923/-10 52.85%  528.5/1000
Stockfish 13 T16   -20  +10/=923/-67 47.15%  471.5/1000

Engine                 Depth       Time   Games     Moves  Average    MIDG   EARLY    ENDG    LATE
Stockfish 13 T32       39.78   26:09:09    1000     76100    1.24    32.34 | 32.95 | 40.31 | 51.18
Stockfish 13 T16       37.19   26:12:55    1000     76061    1.24    30.22 | 30.57 | 37.05 | 48.62

Legend

Depth     : Overall average search depth
Time      : Total time engine used
Moves     : Total moves engine played
Average   : Average time per move in centi-seconds

MIDG      : Average search depth during the opening and middle game
EARLY     : Average search depth during the middle game without Queens, the so called early end game
ENDG      : Average search depth during the end game
LATE      : Average search depth during the late end game
Would be very interesting to see, how much the Elo-spreading increases and the draw-rate decreases, using one of my Anti Draw Openings sets...
Especially my new Drawkiller unbalanced openings produced extremly good results in my testings. The Drawkiller_unbalanced_V4_2mvs set would be very interesting. How much could it shrink this crazy draw-rate in your test of 92.3% ?!?

https://www.sp-cc.de/files/pohl_antidraw_openings.7z

Out of my testing-results:
Average result of 6 classical openings sets in my testings (6x 1000 games = 6000 games):
Score (SF 12 point of view): Score: 53.7% (=+26 Elo), Draws 86.6%
Compared to this:
Drawkiller_unbalanced_V4_2mvs: 1000 (+439,=407,-154), Score: 64.3% (=+102 Elo), Draws 40.7% (!!!)

More testing results and informations on my website:
https://www.sp-cc.de/anti-draw-openings.htm

Or, if you dont like the constructed Drawkiller openings, giving an Unbalanced Human Openings set a try would be also very interesting...
Top
fastgm
Posts: 818
Joined: Mon Aug 19, 2013 6:57 pm

Re: Stockfish 13 - 16 vs 32 threads

Post by fastgm »

Hello Stefan!

Good idea, I'll repeat the test with the book Drawkiller_unbalanced_V4_2mvs

Andreas
Top
User avatar
pohl4711
Posts: 2432
Joined: Sat Sep 03, 2011 7:25 am
Location: Berlin, Germany
Full name: Stefan Pohl

Re: Stockfish 13 - 16 vs 32 threads

Post by pohl4711 »

fastgm wrote: ↑Thu Apr 08, 2021 3:37 pm Hello Stefan!

Good idea, I'll repeat the test with the book Drawkiller_unbalanced_V4_2mvs

Andreas
Cool! Looking forward to the results.
Top
fastgm
Posts: 818
Joined: Mon Aug 19, 2013 6:57 pm

Re: Stockfish 13 - 16 vs 32 threads

Post by fastgm »

Same conditions, 5 min + 3 sec
Hash-Table: 16 GB

Stockfish 13 x64 BMI2 32 threads vs Stockfish 13 x64 BMI2 16 threads, 512 games

+12 Elo

Code: Select all

Stockfish 13 T32   +12  +19/=491/-2 51.66%  264.5/512
Stockfish 13 T16   -12  +2/=491/-19 48.34%  247.5/512

Engine                 Depth       Time   Games     Moves  Average    MIDG   EARLY    ENDG    LATE
Stockfish 13 T32       49.35   66:27:35     512     38546    6.21    41.04 | 42.20 | 50.89 | 59.77
Stockfish 13 T16       46.81   66:21:27     512     38526    6.20    38.57 | 39.29 | 47.87 | 57.69
Games:
http://www.fastgm.de/schach/Stockfish_1 ... 6-300+3.7z
Top
User avatar
pohl4711
Posts: 2432
Joined: Sat Sep 03, 2011 7:25 am
Location: Berlin, Germany
Full name: Stefan Pohl

Re: Stockfish 13 - 16 vs 32 threads

Post by pohl4711 »

fastgm wrote: ↑Mon Apr 12, 2021 3:12 pm Same conditions, 5 min + 3 sec
Hash-Table: 16 GB

Stockfish 13 x64 BMI2 32 threads vs Stockfish 13 x64 BMI2 16 threads, 512 games

+12 Elo

Code: Select all

Stockfish 13 T32   +12  +19/=491/-2 51.66%  264.5/512
Stockfish 13 T16   -12  +2/=491/-19 48.34%  247.5/512

Engine                 Depth       Time   Games     Moves  Average    MIDG   EARLY    ENDG    LATE
Stockfish 13 T32       49.35   66:27:35     512     38546    6.21    41.04 | 42.20 | 50.89 | 59.77
Stockfish 13 T16       46.81   66:21:27     512     38526    6.20    38.57 | 39.29 | 47.87 | 57.69
Games:
http://www.fastgm.de/schach/Stockfish_1 ... 6-300+3.7z
95.9% Draw-rate...=Draw-Death... Ouch. I cant wait for the results, using my new Drawkiller V4 unbalanced openings...
Top
fastgm
Posts: 818
Joined: Mon Aug 19, 2013 6:57 pm

Re: Stockfish 13 - 16 vs 32 threads

Post by fastgm »

I cant wait for the results, using my new Drawkiller V4 unbalanced openings...
The test with the Drawkiller V4 unbalanced openings is currently in progress and should end on Thursday.
Top
fastgm
Posts: 818
Joined: Mon Aug 19, 2013 6:57 pm

Re: Stockfish 13 - 16 vs 32 threads

Post by fastgm »

Same conditions, 60 sec + 0.6 sec
Openings: Drawkiller_balanced_V4_2mvs.pgn (Stefan Pohl)

Stockfish 13 x64 BMI2 32 threads vs Stockfish 13 x64 BMI2 16 threads, 1000 games

+39 Elo

Code: Select all

Stockfish 13 T32   +39  +125/=861/-14 55.55%  555.5/1000
Stockfish 13 T16   -39  +14/=861/-125 44.45%  444.5/1000

Engine                 Depth       Time   Games     Moves  Average    MIDG   EARLY    ENDG    LATE
Stockfish 13 T32       38.11   24:53:44    1000     77083    1.16    32.84 | 33.44 | 37.65 | 48.90
Stockfish 13 T16       35.54   25:00:54    1000     77031    1.17    30.58 | 30.72 | 34.73 | 46.33
Accidentally I took the wrong openings. I have just started the test with Drawkiller_unbalanced_V4_2mvs.
Top
User avatar
pohl4711
Posts: 2432
Joined: Sat Sep 03, 2011 7:25 am
Location: Berlin, Germany
Full name: Stefan Pohl

Re: Stockfish 13 - 16 vs 32 threads

Post by pohl4711 »

fastgm wrote: ↑Wed Apr 14, 2021 7:42 pm Same conditions, 60 sec + 0.6 sec
Openings: Drawkiller_balanced_V4_2mvs.pgn (Stefan Pohl)

Stockfish 13 x64 BMI2 32 threads vs Stockfish 13 x64 BMI2 16 threads, 1000 games

+39 Elo

Code: Select all

Stockfish 13 T32   +39  +125/=861/-14 55.55%  555.5/1000
Stockfish 13 T16   -39  +14/=861/-125 44.45%  444.5/1000

Engine                 Depth       Time   Games     Moves  Average    MIDG   EARLY    ENDG    LATE
Stockfish 13 T32       38.11   24:53:44    1000     77083    1.16    32.84 | 33.44 | 37.65 | 48.90
Stockfish 13 T16       35.54   25:00:54    1000     77031    1.17    30.58 | 30.72 | 34.73 | 46.33
Accidentally I took the wrong openings. I have just started the test with Drawkiller_unbalanced_V4_2mvs.
Nevertheless an interesting testrun with my balanced Drawkiller openings. The Draw-rate is 86.1% here, this is better than the testrun with classical openings, you did first: 94.6%. But a little bit disappointing...SF vs another SF on strong hardware is a very drawish setup... But the really good news is, that the Elo-spreading is nearly doubled with balanced Drawkiller: From +20 Elo in the testrun with classical openings to +39 Elo with Drawkiller balanced. Now, I am curious, how much better (?) the results will be, using my new unbalanced Drawkiller openings!
Thanks for testing!
(Mention, that a doubled Elo-spreading means, only 25% amount of played games are needed to get results/rankings of engines out of the errorbar!)
Top
fastgm
Posts: 818
Joined: Mon Aug 19, 2013 6:57 pm

Re: Stockfish 13 - 16 vs 32 threads

Post by fastgm »

Same conditions, 60 sec + 0.6 sec
Openings: Drawkiller_unbalanced_V4_2mvs.pgn (Stefan Pohl)

Stockfish 13 x64 BMI2 32 threads vs Stockfish 13 x64 BMI2 16 threads, 1000 games

+64 Elo, Draw-rate 51.7%

Code: Select all

Stockfish 13 T32   +64  +333/=517/-150 59.15%  591.5/1000
Stockfish 13 T16   -64  +150/=517/-333 40.85%  408.5/1000

Engine                 Depth       Time   Games     Moves  Average    MIDG   EARLY    ENDG    LATE
Stockfish 13 T32       33.78   27:22:13    1000     90536    1.09    29.52 | 29.52 | 33.00 | 44.27
Stockfish 13 T16       31.51   27:31:17    1000     90425    1.10    27.69 | 27.50 | 30.38 | 41.84
Top

Return to “Computer Chess Club: Tournaments and Matches”