Monday, March 26, 2007

Start your engines

I recently fulfilled a promise made a couple of months ago and sent off 34 euros to IM and chess programmer Vasik Rajlich for a copy of Rybka 2.3. I had hoped to be writing that this new release exceeds all expectations but this has not proven to be so, at least as far as my copy is concerned. The problem is that Rybka 2.3 runs much more slowly than other programs and this not only affects its playing strength but also causes strange lag effects in the board display. A posting on the Rybka Forum indicates a hashtable bug that is due to be corrected very shortly through a beta release. That's fine with me as long as it doesn't cost anything.

In order to compare Rybka 2.3 with my other playing programs I decided to assign a position from the game Szabo-Bisguier, Buenos Aires 1955, which is shown in the diagram. Szabo concluded matters by force in only three more moves: 1.Bxh7+ Kxh7 2.Qh3+ Kg8 3.Rg4! and Black resigned because he cannot prevent Rh4-h8 mate. His king is blocked in by his own pieces and any move of the f-pawn is met by g5-g6. This is a good test position for humans because the key move (3.Rg4) is slightly counterintuitive; it is usually the queen that is on the sharp end of h-file attacks.

Unlike human players, computers are tactically perfect within a short range and are thought to have few biases so one would expect them to solve this position fairly quickly. I have a number of UCI engines that run under the Fritz interface and I gave them all the same position in turn. Of course they all solved it but they took varying amounts of time to do so. Here are the results:

Rybka 1.0 beta: 2 seconds
Fritz 8: 17 seconds
Pro Deo 1.1: 60 seconds
Fruit 2.1: 62 seconds
Toga II 1.2.1: 68 seconds
Rybka 2.3: 72 seconds
Crafty 19.15: 189 seconds

Yes, a little joke at the end... I wonder what a match between Rybka 1.0 and Crafty 19.15 would look like? In any case I think you can sense my concern over spending 34 euros on Rybka: it was outperformed by the free programs Pro Deo, Fruit and Toga, and badly outperformed by Fritz 8. They were all running on an AMD Athlon 2600+ with 64 MB hashtables.

Any and all feedback is welcome on this topic.

1 comment:

Ben Matheson said...

Not a big deal... but I think your second move should be ... K-g8 and not K-h8. I'm sure it must just be a typo. :)

PS. I enjoy your updates.

About Me

My photo
Port Coquitlam, British Columbia, Canada
National master (Canada) since 1984. B.C. Champion 1977 and 1984. Runner-up 1991 and 2002. B.C. Open Champion 1972 and 1982. B.C. U/14 Champion 1964-65-66. Mikhail Botvinnik once wrote that publishing your analytical work forces you to be accurate because it exposes you to criticism. Hence this blog.