Public benchmarks

[bedouin]

Are there any public benchmarks available? I have fritz9_benchmarks bookmarked. Any other links some of you know?

[Dann Corbit]

http://www.xs4all.nl/~mlngveld/craftybench/index.html
http://www.sedatchess.com/hardwares.html
http://www.geocities.com/sedatchess/hardware.html
http://www.beepworld.de/members39/compu ... smarks.htm
http://www.vlasak.biz/evcpu.htm
http://www.beepworld.de/members38/lchri ... -chess.htm

I think that crafty is part of SPEC now, so you should be able to get crafty NPS figures for just about every CPU made

[Sedat Canbaz]

Here is another fritz benchmark site:
http://www.jens.tauchclub-krems.at/dive ... marks.html

Best,
Sedat

[bedouin]

Here is another fritz benchmark site:
http://www.jens.tauchclub-krems.at/dive ... marks.html

Best,
Sedat

That appears to be the same site I linked to above.

[bedouin]

Its interesting to see the OverClocked QX6600 Quad Core achieving

Code: Select all

Intel QX6600 Quad Core OC  	4 x 3600  	21,54  	10337

while the QX6700 Quad Core OC

Code: Select all

Intel QX6700  Quad Core OC  	4 x 3200  	   	8518

Is that an error perhaps?

[ernest]

No error:
the QX6600 is overclocked to 3600 MHz
the QX6700 is (less) overclocked to 3200 MHz

[Vempele]

That still leaves 3600/3200 < 10337/8518. You'd expect QX6700 to be at least equally fast per MHz, right? Unless there's reason to believe QX6600 is indeed faster than QX6700, the benchmark used is flawed.

They probably just ran the benchmark once per machine. Perfectly fine on single cores (as you'll get the exact same result every time), but non-deterministic on anything else.

[ernest]

That still leaves 3600/3200 < 10337/8518. You'd expect QX6700 to be at least equally fast per MHz, right? Unless there's reason to believe QX6600 is indeed faster than QX6700, the benchmark used is flawed.

They probably just ran the benchmark once per machine. Perfectly fine on single cores (as you'll get the exact same result every time), but non-deterministic on anything else.

You are right, 8518 is not very good. Maybe the memory settings were not so good?
On my own E6600 @ 3000 MHz, with excellent memory and 4-4-4-12 settings, I obtained 4350, which is about the best ratio of the list (1.45 per MHz, for 2-core)
http://www.jens.tauchclub-krems.at/dive ... marks.html
But you are wrong when you speak of non-deterministic: the nodes-numbers in the "Fritz 9 Schachbenchmarks" are deterministic (quad is 2x 2-core and 2-core is 2x single-core).This has nothing to do with real chess analysis, where 2-core is approx 1.7x single-core and where depths, PVs and solution times are indeed non deterministic for more than 1-core.
It can be noted that the 10337 is indeed very good, it corresponds, for 2-core scale, to 10337/2/3600 = 1.436 !
The guy used watercooling, see http://www.jens.tauchclub-krems.at/dive ... ch/rob.jpg

[bedouin]

Fritz with its 8 cores achieved 12751 kN/s while the 16 core Junior achieved 25690 kN/s in the President's cup (see http://www.chessbase.com/newsdetail.asp?newsid=3920). I don't believe that the alleged machines at Mission Control's disposal have ever produced a public benchmark. (Alleged is the correct word to use because Mission Control played the flyingfatman game and said his machines foresaw the defeat a long way off but he allowed an exchange sac when a normal rybka shows its good for his opponent.)

[Nid Hogge]

Yes, But you can't compare them directly. It wasn't running FCB, They just captured a moment in the game that could cary and depend heavily on the current position.
Deep Junior was always known for it's crazy Kn/S. It's probably the fastest calculator out there. the 16 cores must have helped it as well.

I don't know about Mission Control as well. All Windows versions can only support 64 Cores.

So to use more cores than that he must have used a Linux distro.
And we all know Chessbase\Playchess interface does not run on Linux, So he couldn't have played online there.
Rybka doesn't run on linux as well.

So even if he's honest, the max # of Cores he could effectivly use on Windows is 64 Which is not bad at all anyway.

Theoretically he could analyse with 128 Cores on Linux and Arena and run on centaur mode, But I don't know how many programs there are out there that support linux and are able to utilitize so many Cores.

PS Chinese Overclocker VictorWang already broke 15k and 30 RS in Fritz Chess Benchmark:



I guess he has the WR. Wonder if the list is still updated.