Rybka 2.3.2a vs. Fritz 11 and vs Shredder 11

[Christopher M]

Rybka 2.3.2a vs. Fritz 11 and vs Shredder 11, 100 Blitz games each

100 games for each engine match:

Rybka 2.3.2a - Fritz 11, Blitz 5min + 5sec (total time per side ~ 10+ minutes)

1 Rybka 2.3.2a 32-bit +23/-14/=63 54.50% 54.5/100
2 Fritz 11 +14/-23/=63 45.50% 45.5/100

Rybka 2.3.2a - Shredder 11, Blitz:5min + 5sec (total time per side ~10+minutes)

1 Rybka 2.3.2a 32-bit +60/-13/=27 73.50% 73.5/100
2 Shredder 11 +13/-60/=27 26.50% 26.5/100


Rybka still on top, of course, but Fritz 11 is clearly making progress compared to older versions.

AMD Athlon 3000+, 768MB RAM, 132MB Hash Tables each
Fritz 11 GUI
Each engine used their own books: RybkaII.ctg, Shredder11.ctg, and Fritz11.ctg
Default settings used for each of the engines.
All 3, 4 and 5 piece EGTBs

An interesting aspect of the Rybka - Fritz 11 match is the huge difference in speed of the two engines. The speed and depth search in plys of each engine is given in the score of each game. Typical numbers here are Fritz 11 - 19.3 ply, speed 597 kN/s, Rybka 17.3 ply, speed 38 kN/s. Thus, even though Fritz 11 was looking a full move ahead of Rybka (2 play = one move by white and one by black), and searching almost 16 times faster, Rybka won the 100 game match handily. Speed and search depth are clearly not the determinative factors here, thus it must be chess knowledge.

A number of the games did not end in a decision, just a “line” per the GUI, but all of such games only after 57-100+ moves, that is the game did not crash early. These were easy to adjudicate. First, a majority of them were clearly drawn games as material was even, and each of the combatant engines scored the ending position no more than a half pawn in favor of one side, and most of the rest were clear winners for one side or the other: material difference of three plus points (pawns), and ending position scores of the two combatant engines in agreement favoring one side or the other by 2.5 - 10+ points. The adjudications were confirmed with a third engine, here Hiarcs 11.2 MP. No win was awarded for a “line” game unless there was a material advantage for one side, and all three engines agreed that the final position score favored that side by 1.5+ points. Of the latter, closer games, there were only three among the 200 played.

Games available from me by request in cbv (Chessbase) or PGN zipped at camorglaw@roadrunner.com.

Christopher M

[Eizenhammer]

Typical numbers here are Fritz 11 - 19.3 ply, speed 597 kN/s, Rybka 17.3 ply, speed 38 kN/s. Thus, even though Fritz 11 was looking a full move ahead of Rybka (2 play = one move by white and one by black), and searching almost 16 times faster, Rybka won the 100 game match handily. Speed and search depth are clearly not the determinative factors here, thus it must be chess knowledge.Christopher M

Well, you may not know it, but it is pretty well known that rybkas output is manipulated, (or by different words: it counts in a different way), so that you have to add something like 2 to the search depth and multiply the number of knodes by a factor of something like x (maybe 600/40) .

You are free to draw conclusions based on the new numbers.

regards
Peter

[Ovyron]

you have to add something like 2 to the search depth

I think that it's actually 3 plies so Rybka would be at 20.3 ply, one half move faster than Fritz. Also, the nodes per second have been adjusted to show the effective speed improvement (according to Vas) of going from 1 core to more cores, or from 32bit to 64bit. That's their only purpose, to be compared with other Rybkas on different setups, but comparing them to other engine's nps just makes no sense, and you would need a different formula for converting those to a more familiar node counting, depending on the number of cores and whether the Rybka is 32 or 64bit.

[Christopher M]

Peter and Ulysses,

Thanks for the info.

Christopher