Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Crafty Binaries/CPU sepcific versions - Test results & Questions

Author: Mike Byrne

Date: 14:44:30 07/03/02

Go up one level in this thread


Hi Shaun,

Those are interesting stats, I took a  particular interest, since I compiled the
latest version of crafty on Bob Hyatt's site.

I'm not a programmer by trade and I am sure I have compiled far fewer "crafty's"
than Dan Corbitt.  My guess is that Dan is a "little" bit (meaning = "a lot
more") knowledgeable in what optimaztions work best for Crafty.  I am not
surprised at all at the differences.  Also, if you are using a P4, my executable
"wcrafty" has a slight penalty as it checks to see what processor it is running
on so that it will know which optimaztions will work.
The P4 version will work better than than a non P4 version on a P4 machine.

I really would be interested in how the P4 version stacks up against Dan's.

Dan,

I'm interested in your makefile if you can send it to me.  Also I have been
unable to you site over the last two months - not sure what the deal is there -
so I can't run these tests myself.

Thanks,

Mike


On July 03, 2002 at 10:43:55, Shaun Brewer wrote:

>I have been looking at some different binaries for crafty 18.15 - I was looking
>for CPU optimized versions. I have run tests on the binaries I have found - see
>at the end of this note.
>
>A couple of questions have occurred to me:
>
>Are there other sites that provide various optimized binaries?
>
>Given there does seem to be a significant difference in performance of these
>binaries do other programs have CPU optimized executables?
>
>Results:
>--------
>
>Bench results:
>
>Each bench was run three times and the best result taken - normally very little
>difference but attempt to remove any discrepancy due to background tasks
>
>Bench results on Athlon (TBird) 850mhz
>
>wcrafty-18.15 - from Robert Hyatt’s ftp site
>
>Total nodes: 97487547
>Raw nodes per second: 451331
>Total elapsed time: 216
>SMP time-to-ply measurement: 2.962963
>
>Crafty18-15 - from Dann Corbit’s ftp site
>
>Total nodes: 97487547
>Raw nodes per second: 480234
>Total elapsed time: 203
>SMP time-to-ply measurement: 3.152709
>
>Crafty-1815 - from Dann Corbit’s ftp site
>
>Total nodes: 111139080 - ??? different total nodes
>Raw nodes per second: 479047
>Total elapsed time: 232
>SMP time-to-ply measurement: 2.758621
>
>c1815-k7mmx - from ftp://speedycpu.dyndns.org/pub/crafty/
>
>Total nodes: 111139080
>Raw nodes per second: 479047
>Total elapsed time: 232
>SMP time-to-ply measurement: 2.758621
>
>Bench results on PIII 1132mhz
>
>wcrafty-18.15 - from Robert Hyatt’s ftp site
>
>Total nodes: 97487547
>Raw nodes per second: 590833
>Total elapsed time: 165
>SMP time-to-ply measurement: 3.878788
>
>Crafty18-15 - from Dann Corbit’s ftp site
>
>Total nodes: 97487547
>Raw nodes per second: 617009
>Total elapsed time: 158
>SMP time-to-ply measurement: 4.050633
>Crafty-1815 - from Dann Corbit’s ftp site
>
>Total nodes: 111139080
>Raw nodes per second: 614028
>Total elapsed time: 181
>SMP time-to-ply measurement: 3.535912
>
>c1815-k7mmx - from ftp://speedycpu.dyndns.org/pub/crafty/
>
>Total nodes: 111139080
>Raw nodes per second: 669512
>Total elapsed time: 166
>SMP time-to-ply measurement: 3.855422
>
>
>Match Results
>-------------
>
>Match results all on Athlon (TBird) 850mhz
>
>All games 2min + 2sec - If anyone wants me to post PGN I can - or email:
>
>Stats produced with elostat_11b base elo 2400 (I had to choose something)
>
>wcrafty-18.15 v Crafty-1815
>---------------------------
>
>1st 100 games
>
>    Program                            Score     %    Av.Op.  Elo    +   -
>Draws
>
>  1 Crafty-1815                    :  57.0/100  57.0   2376   2424   64  45
>46.0 %
>  2 wcrafty-18.15                  :  43.0/100  43.0   2424   2376   45  64
>46.0 %
>
>1st 200 games
>
>    Program                            Score     %    Av.Op.  Elo    +   -
>Draws
>
>  1 Crafty-1815                    : 109.0/200  54.5   2384   2416   46  27
>55.0 %
>  2 wcrafty-18.15                  :  91.0/200  45.5   2416   2384   27  46
>55.0 %
>
>
>1st 300 games
>
>    Program                            Score     %    Av.Op.  Elo    +   -
>Draws
>
>  1 Crafty-1815                    : 157.0/300  52.3   2392   2408   39  21
>58.7 %
>  2 wcrafty-18.15                  : 143.0/300  47.7   2408   2392   21  39
>58.7 %
>
>
>All 362 games
>
>    Program                          Elo    +   -   Games   Score   Av.Op.
>Draws
>
>  1 Crafty-1815                    : 2406   36  19   362    51.7 %   2394   58.0
>%
>  2 wcrafty-18.15                  : 2394   19  36   362    48.3 %   2406   58.0
>%
>
>
>wcrafty-18.15 v Crafty18-15
>---------------------------
>
>1st 100 games
>
>    Program                            Score     %    Av.Op.  Elo    +   -
>Draws
>
>  1 Crafty18-15                    :  55.0/100  55.0   2383   2417   65  34
>62.0 %
>  2 wcrafty-18.15                  :  45.0/100  45.0   2417   2383   34  65
>62.0 %
>
>1st 200 games
>
>    Program                            Score     %    Av.Op.  Elo    +   -
>Draws
>
>  1 Crafty18-15                    : 108.5/200  54.2   2385   2415   46  26
>57.5 %
>  2 wcrafty-18.15                  :  91.5/200  45.8   2415   2385   26  46
>57.5 %
>
>1st 300 games
>
>    Program                            Score     %    Av.Op.  Elo    +   -
>Draws
>
>  1 Crafty18-15                    : 159.0/300  53.0   2390   2410   38  20
>59.3 %
>  2 wcrafty-18.15                  : 141.0/300  47.0   2410   2390   20  38
>59.3 %
>
>All 360 games
>
>    Program                            Score     %    Av.Op.  Elo    +   -
>Draws
>
>  1 Crafty18-15                    : 189.5/360  52.6   2391   2409   35  18
>61.4 %
>  2 wcrafty-18.15                  : 170.5/360  47.4   2409   2391   18  35
>61.4 %
>
>
>Rating difference appears to drop over time due to learning?
>
>Crafty18-15 v Crafty-1815
>
>to follow
>
>wcrafty-18.15 v c1815-k7mmx
>
>to follow



This page took 0 seconds to execute

Last modified: Thu, 15 Apr 21 08:11:13 -0700

Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.