Komodo 3 SSE42 running for the IPON

[IWB]

As usuall here: http://www.inwoba.de

Let's see where it ends ...

Have fun
Ingo

[Leto]

Thanks. 2976 after 208 games, that would put it 22 elo above Rybka 4, 35 elo below Houdini 1.5a. Still very early though.

[Leto]

2973 after 716 games. It now has a positive score against Houdini 1.5a (54%), but a negative score against Rybka 4 (37%). Interesting.

[Vinvin]

2973 after 716 games. It now has a positive score against Houdini 1.5a (54%), but a negative score against Rybka 4 (37%). Interesting.

This strange phenomenon is confirmed by the Clemens Keck's match :
http://www.clemens-keck.de/livegames4/komodo3sse.html
Komodo 3 scores 47% against Rybka and 55% against Houdini

Code: Select all

Komodo64 3 sse - Deep Rybka 4.1 SSE42 x64 x1 (2938)		18.0	-	20.0		47.37%		Perf=2920
Komodo64 3 sse - Houdini 1.5a x64 x1 (3009)		20.5	-	16.5		55.41%		Perf=3046

[Laskos]

2973 after 716 games. It now has a positive score against Houdini 1.5a (54%), but a negative score against Rybka 4 (37%). Interesting.

This strange phenomenon is confirmed by the Clemens Keck's match :
http://www.clemens-keck.de/livegames4/komodo3sse.html
Komodo 3 scores 47% against Rybka and 55% against Houdini

Code: Select all

Komodo64 3 sse - Deep Rybka 4.1 SSE42 x64 x1 (2938)		18.0	-	20.0		47.37%		Perf=2920
Komodo64 3 sse - Houdini 1.5a x64 x1 (3009)		20.5	-	16.5		55.41%		Perf=3046

Too few games.

[Leto]

2965 after 1130 games. 53% against Houdini 1.5a, 42.5% against Rybka 4., 51% against Critter 1.2, and 57% against Stockfish 2.1.1 JA.

[Dr.Wael Deeb]

2965 after 1130 games. 53% against Houdini 1.5a, 42.5% against Rybka 4., 51% against Critter 1.2, and 57% against Stockfish 2.1.1 JA.

Soon Robert will kiss his chess throne goodbye....real soon

[Laskos]

2965 after 1130 games. 53% against Houdini 1.5a, 42.5% against Rybka 4., 51% against Critter 1.2, and 57% against Stockfish 2.1.1 JA.

Soon Robert will kiss his chess throne goodbye....real soon

Why? It's only +12 now for Kom 3, and the new Houdini could be +50 to the current advantage. There was a brick-wall in progress in 2003-4, couldn't it happen now?

Edit: Besides that, all the best engines apart from Houdini are only marginally better than the open source, 2 years old Ippolit.

[lkaufman]

Note that as of this writing the Keck list shows +33 over Komodo 2.03. I think that the explanation of the discrepancy, aside from sample error, is that 2.03 was compiled with optimization for AMD (which IPON uses), and 3.00 was not. I expect that the CCRL and CEGT testing will confirm +30 for 3.00, especially considering time control improvements that do not show up on IPON.
I don't expect that we will be able to maintain our 4 elo per week pace so far this year, even after a break to go MP. If we lose Sept. due to MP work and then gain 3 elo per week, which seems reasonable, we should catch the current Houdini in January. Just an educated guess.

[Laskos]

Note that as of this writing the Keck list shows +33 over Komodo 2.03. I think that the explanation of the discrepancy, aside from sample error, is that 2.03 was compiled with optimization for AMD (which IPON uses), and 3.00 was not. I expect that the CCRL and CEGT testing will confirm +30 for 3.00, especially considering time control improvements that do not show up on IPON.
I don't expect that we will be able to maintain our 4 elo per week pace so far this year, even after a break to go MP. If we lose Sept. due to MP work and then gain 3 elo per week, which seems reasonable, we should catch the current Houdini in January. Just an educated guess.

That would explain my AMD results. I am running now a test at twice TC (3s + 0.1s), to the average depth ~11, the result in self-testing K3 - K2.03 seems again pretty close to equal (still too few games).

A rate of even 2 points per week is excellent and twice the yearly average for engines, so I hope you could keep it.

Kai