Rybka 3 vs Stockfish 1.6s on Octa 4ghz 8 core

[M ANSARI]

5 min + 1 sec games using Perfect 12 books truncated at 10 move

Stockfish 1.6s_Rybka 3_5_1-2 2009


1 Rybka 3 +37/=51/-12 62.50% 62.5/100
2 Stockfish 1.6s 64bit +12/=51/-37 37.50% 37.5/100



Rybka 3 running at default since I forgot to put contempt at zero. That might have actually improved scores although when I quickly looked at the games there is one loss by Rybka due to high contempt. This is probably offset though by less draws due to the contempt since it seems R3 is still quite a bit stronger. Maybe someone else can look at the games and see if they do find out something. I would say that that Stockfish 1.6s is the strongest non Rybka engine. Of course I consider Rybka clones as also Rybka engines.

Forget the games as I can't seem to find the attach button.

[Tord Romstad]

Stockfish is not likely to perform well on an 8 core machine. Optimizing the search for 8 threads requires lots of testing and tuning on an 8 core machine, and Stockfish's parallel search was developed and tested on an ordinary dual core computer. I've never even tried running the program on a quad-core box. It's quite strange that it performs as well as it does on quad-cores on the CCRL and CEGT rating lists.

It would surprise me if Stockfish played any better at all with 8 threads than with 4. I guess that the optimum number of threads is somewhere between 4 and 6, but it is impossible to be sure without lots of testing.

As long as virtually everyone is still using single- or dual-core computers, I don't worry much about this.

[Suj]

Changing the maximum number of threads per split point to 9 helps.

Also to increase number of threads to 8.

This worked much better when I tested stockfish on 8 to 32 cores.

[Dr.Ex]

The DC-compile Stockfish32.exe "slightly smoother" seems to work just great on my Quad Core with WXP 32-bit and on my Dual Core WXP 32-bit Laptop.
A 20 game match 4+2 against Rybka 3 MP 32-bit on Dual Core with ponder off ended 10-10 (+3 =14 -3), on Quad Core in a 5+1 ponder off match Stockfish is even 12.5-7.5 (+8 =9 -3) in the lead at the moment.
So I would say Stockfish 1.6s DC 32-bit is slightly stronger than the best _official_ 32-bit Rybka, according to you

[alpha123]

In a 3/0 Nunn match on my quad, Stockfish 1.6 JA 64bit 4-CPU drew Rybka 3 1-cpu x64 +8/=4/-8, which would suggest that Stockfish is still weaker than Rybka . Games here.

NOTE: This was normal 1.6, not 1.6s.

Peter

[beachknight]

5 min + 1 sec games using Perfect 12 books truncated at 10 move

Stockfish 1.6s_Rybka 3_5_1-2 2009


1 Rybka 3 +37/=51/-12 62.50% 62.5/100
2 Stockfish 1.6s 64bit +12/=51/-37 37.50% 37.5/100



Rybka 3 running at default since I forgot to put contempt at zero. That might have actually improved scores although when I quickly looked at the games there is one loss by Rybka due to high contempt. This is probably offset though by less draws due to the contempt since it seems R3 is still quite a bit stronger. Maybe someone else can look at the games and see if they do find out something. I would say that that Stockfish 1.6s is the strongest non Rybka engine. Of course I consider Rybka clones as also Rybka engines.

Forget the games as I can't seem to find the attach button.

Another blitz tournament is underway: Stockfish Special:

Code: Select all

    Engine                   Score StDeStRyNaStStStThStRoRoGlGrTo    S-B
01: Stockfish 1.6.0 JA 2     7.5/11 · = = 0 1 =   =   1 = 1 1   1   35,25
02: Deep Shredder 12         7.0/10 = · 1 =   = = 1   =   1 =   1   36,00
03: Stockfish 1.6.0 DC       7.0/12 = 0 · 1 = = = = =   = =   1 1   34,25
04: Rybka 3                  7.0/10 1 = 0 ·     = 1 = =   1 1   1   33,75
05: Naum 4                   6.5/10 0   =   · = = 1 = 1 =   1 1     30,50
06: Stockfish 1.6.0 JA 3     6.0/11 = = =   = · = 0 1 = = =   1     32,00
07: Stockfish 1.5.1 JA Intel 6.0/11   = = = = = ·   = = 0   1 = 1   27,25
08: Stockfish 1.6.0 DC 2     6.0/12 = 0 = 0 0 1   ·   0 1 = = 1 1   25,25
09: Thinker 5.4Di            5.5/10     = = = 0 =   · = 1   0 1 1   24,25
10: Stockfish 1.6.0 JA       5.0/11 0 =   = 0 = = 1 = · 1 0   =     27,50
11: RobboLito 0.085g3        4.5/10 =   =   = = 1 0 0 0 · =   1     24,25
12: RobboLito 0.085e4        4.5/10 0 0 = 0   =   =   1 = · =   1   19,50
13: Glaurung 2.1 AMD64       2.5/10 0 =   0 0   0 = 1     = · 0 0   14,25
14: Grapefruit 1.0 DM        2.5/10     0   0 0 = 0 0 = 0   1 · =    8,75
15: Toga II 2.0 SE w32       1.5/10 0 0 0 0     0 0 0     0 1 = ·    3,75

79 of 105 games played
Name of the tournament: CET2402o
Site/ Country: CET24_Antalya_TUR, Türkiye
Level: Blitz 2/2
Hardware: Intel(R) Core(TM)2 Quad CPU    Q6600  @ 2.40GHz4x  2400 MHz with 2.048 MB Memory
Operating system: Microsoft Windows XP 64 Bit Professional Service Pack 2 (Build 3790)

[ernest]

using Perfect 12 books truncated at 10 move

Hi Majd,
Is this Sedat's Perfect v12.ctg (10/10/2006) ?
or some other (to be downloaded where?)?

[M ANSARI]

Yes that is Sedat's book which you can download at his site.

[alpha123]

11: RobboLito 0.085g3 4.5/10 = = = = 1 0 0 0 · = 1 24,25

I have a Q8300 @ 2.50gHz, and RobboLito performs much better than that.
Why does it seem to be so bad on your machine?!?!

Peter

[Dann Corbit]

11: RobboLito 0.085g3 4.5/10 = = = = 1 0 0 0 · = 1 24,25

I have a Q8300 @ 2.50gHz, and RobboLito performs much better than that.
Why does it seem to be so bad on your machine?!?!

Peter

Probably because he has 8 cores and Robbo engines use only 1.