SPCC: Testrun of CFish 12 finished

[pohl4711]

AB-testrun of CFish 12 avx2 finished.

https://www.sp-cc.de

(Perhaps you have to clear your browsercache or reload the website)

[ernest]

Hi Stefan,
Where to download this Cfish 12 ?
TIA

[sarona]

You can find it here https://github.com/syzygy1/Cfish/releases

[ernest]

Thanks !

[Jouni]

I check speed up to SF12 abrok in Haswell CPU. SF (default) bench is useless as I says +2% only. With 4 cores +9%. But real games says +25% :

Code: Select all

 1.  Cfish 12 64 BMI2 NUMA    	6.0/10	2-0-8  	(L: m=0 t=0 i=0 a=0)	(D: r=3 i=2 f=0 s=0 a=3)	(tpm=1274.3 d=28.04 nps=1614728)
 2.  Stockfish 12             	4.0/10	0-2-8  	(L: m=0 t=0 i=0 a=2)	(D: r=3 i=2 f=0 s=0 a=3)	(tpm=1322.7 d=27.32 nps=1285578)

I assume default mode is hybrid use of NNUE in both.

[MMarco]

I check speed up to SF12 abrok in Haswell CPU. SF (default) bench is useless as I says +2% only. With 4 cores +9%. But real games says +25% :

Code: Select all

 1.  Cfish 12 64 BMI2 NUMA    	6.0/10	2-0-8  	(L: m=0 t=0 i=0 a=0)	(D: r=3 i=2 f=0 s=0 a=3)	(tpm=1274.3 d=28.04 nps=1614728)
 2.  Stockfish 12             	4.0/10	0-2-8  	(L: m=0 t=0 i=0 a=2)	(D: r=3 i=2 f=0 s=0 a=3)	(tpm=1322.7 d=27.32 nps=1285578)

I assume default mode is hybrid use of NNUE in both.

On my system (Ryzen 4900H) SF nps are higher with the <bench> command (which is actually equivalen to <bench 16 1 13>). When I increase depth (i.e <bench 64 1 20>) CF start doing better up to 21-22%. In game play, CFish is around 35% faster than SF... the bench results is somewhat unreliable.