Komodo 3 SSE42 running for the IPON

[IWB]

Komodo 3 finished it's initial run.

A new second rank in the IPON list!

Thanks for your attention
Ingo

PS: Komodo missed the best day in page hits by 200 hits. Stockfish is still the leader in my internal popularity ranking

[Laskos]

Komodo 3 finished it's initial run.

A new second rank in the IPON list!

Thanks for your attention
Ingo

PS: Komodo missed the best day in page hits by 200 hits. Stockfish is still the leader in my internal popularity ranking

Thanks, so it's +13 Elo points by Kom 3 compared to Kom 2.03. I could confirm this result (on an AMD) at ultra-fast control 3s + 0.1s

Code: Select all

    Program                            Score       %     Elo    +   -    Draws

  1 Komodo64 3                     : 2880.0/5570  51.7   3006    7   7   36.0 %
  2 Komodo64 2.03 JA               : 2690.0/5570  48.3   2994    7   7   36.0 %

12 +/- 7 Elo points improvement 95% confidence.

Kai

[IWB]

Hi

I will play a few more games (e.g. Fritz) but I doubt a big change in rating. The good side is, that the result now might be in line with the other lists, the bad thing is that the compile now is prefering Intel again ... The AMDs could do much better if the compilers would take care (see 2.03 DC!). But with the current situation AMD has to be better than Intel to get equal results ... and I fear this is the case not only for chess.

Bye
Ingo

[Albert Silver]

Hi

I will play a few more games (e.g. Fritz) but I doubt a big change in rating. The good side is, that the result now might be in line with the other lists, the bad thing is that the compile now is prefering Intel again ... The AMDs could do much better if the compilers would take care (see 2.03 DC!). But with the current situation AMD has to be better than Intel to get equal results ... and I fear this is the case not only for chess.

Bye
Ingo

Out of curiosity, does it make a big diff when Abblett makes his compiles? I don't mean in terms of CPU compatibility, but in performance.

[Eelco de Groot]

It is just my guess but I believe that Jim prefers to make builds that run on most machines, so especially the popcount instruction for faster counting the number of bits in a bitboard is out. Richard Vida stated that looking at the sources of a previous version I think, Komodo uses popcount a lot, more than other programs, that is kind of inevitable if you have a lot of knowledge in the evaluation function of a bitboard program.

That is why Dann's compile for the i7 should be as fast as Jim's compiles, I am not sure it also runs directly on i5 but I suspect it does?. And it depends very much on the engine too what the final differences are. If Komodo prefers Intel processors it may be partly because Dann is using his Intel compiler (which unfortunately really favours Intel) or it may be that Komodo was tested more on Intel. I believe Larry has a 12 core Intel

Larry: It's an I7 but just 2.2 Ghz.

Larry: It's two six core I7s in one machine. Maybe they are not technically I7s, but they are basically the same, just designed for this purpose.

Lukas:I guess it's 2x Xeon Westmere L5640

so it figures that both testing and compiling will favour the i7, to a lesser extent the i5.

[Eelco de Groot]

Edit: Maybe I am mistaken about this, that the general version can't use hardware popcount instruction of i7 and i5, it is better to ask Don about this.

Eelco