Page 1 of 4

New CFish Windows binary (from today's code)

Posted: Thu Oct 13, 2016 12:17 am
by Dann Corbit
The "Cfish-vanilla.exe" binary is fastest for me.
https://www.dropbox.com/s/0zfnehxzq7ery ... er.7z?dl=0

Re: New CFish Windows binary (from today's code)

Posted: Fri Oct 14, 2016 7:47 am
by giovanni
Thanks, Dann. Do you have any speed comparison data with the original program?

Re: New CFish Windows binary (from today's code)

Posted: Fri Oct 14, 2016 1:30 pm
by Dann Corbit
giovanni wrote:Thanks, Dann. Do you have any speed comparison data with the original program?
It is hard to make a clear standardized comparison because the bench changed and Cfish does not have the new positions.

It is a lot faster for me, depending on the position, up to twice as fast.

For me, the "vanilla" version is fastest.

Re: New CFish Windows binary (from today's code)

Posted: Fri Oct 14, 2016 2:12 pm
by PaulieD
The vanilla or modern binary is not working on my popcount system

Re: New CFish Windows binary (from today's code)

Posted: Fri Oct 14, 2016 3:37 pm
by Carlos777
Hi Paul,

Have you tried to compile it?

I just go to the folder where the source is, open a windows console, type "make clean" and then type "make cfish" (without quotes)

It worked for me.

Regards,
Carlos

Re: New CFish Windows binary (from today's code)

Posted: Fri Oct 14, 2016 4:55 pm
by shrapnel
Dann Corbit wrote:The "Cfish-vanilla.exe" binary is fastest for me.
https://www.dropbox.com/s/0zfnehxzq7ery ... er.7z?dl=0
Modern version is really fast and good on my System.
How can I get a BMI version ?

Re: New CFish Windows binary (from today's code)

Posted: Fri Oct 14, 2016 4:59 pm
by Jouni
"twice as fast"? According to "bench" CFish is 12,5% faster than abrok compile. But for some reason Cfish loses almost 10% of games in LittleBlitzer, when time control is 10s + 0,1s! SF and asmFish only may be once in 500 games normally. Why?

Re: New CFish Windows binary (from today's code)

Posted: Fri Oct 14, 2016 5:41 pm
by syzygy
Jouni wrote:"twice as fast"? According to "bench" CFish is 12,5% faster than abrok compile. But for some reason Cfish loses almost 10% of games in LittleBlitzer, when time control is 10s + 0,1s! SF and asmFish only may be once in 500 games normally. Why?
Do you mean it loses on time? It should not behave differently in this respect from Stockfish, but this is not something that I have tested very well.

If the losses are not time losses but crashes, then I might be about to fix it (yesterday I noticed a problem which I will look into as soon as I have time).

Re: New CFish Windows binary (from today's code)

Posted: Fri Oct 14, 2016 8:08 pm
by Jouni
Yes I mean time losses as seen in this log:

Code: Select all

Games Completed = 76 of 1000 (Avg game length = 28.774 sec)
Settings = Gauntlet/256MB/10000ms+100ms/M 450cp for 4 moves, D 100 moves/EPD:10moves_SALC_500.epd(500)
Time = 567 sec elapsed, 6892 sec remaining
 1.  Cfish 121016 64 POPCNT   	45.5/76	33-18-25  	(L: m=0 t=6 i=0 a=12)	(D: r=18 i=0 f=0 s=0 a=7)	(tpm=271.9 d=17.15 nps=2092287)
 2.  Komodo 10 64-bit         	30.5/76	18-33-25  	(L: m=0 t=0 i=0 a=33)	(D: r=18 i=0 f=0 s=0 a=7)	(tpm=265.3 d=15.68 nps=1913349)
If I raise inc to 200ms no help!

Re: New CFish Windows binary (from today's code)

Posted: Sat Oct 15, 2016 6:29 am
by Jouni
Average time per move (tpm) is for Cfish clearly bigger than Komodo's. Normal SF (abrok) uses less time always.

Code: Select all

Games Completed = 46 of 1000 (Avg game length = 27.040 sec)
Settings = Gauntlet/256MB/10000ms+100ms/M 450cp for 4 moves, D 100 moves/EPD:10moves_SALC_500.epd(500)
Time = 326 sec elapsed, 6758 sec remaining
 1.  Stockfish 260916 64 BMI2 	30.0/46	21-7-18  	(L: m=0 t=0 i=0 a=7)	(D: r=13 i=3 f=0 s=0 a=2)	(tpm=255.7 d=17.04 nps=1805392)
 2.  Komodo 10 64-bit         	16.0/46	7-21-18  	(L: m=0 t=0 i=0 a=21)	(D: r=13 i=3 f=0 s=0 a=2)	(tpm=267.2 d=15.86 nps=1936614)