Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Cilkchess

Author: Vincent Diepeveen

Date: 04:48:42 02/26/02

Go up one level in this thread


On February 25, 2002 at 20:27:56, K. Burcham wrote:

>
>I know most here are familiar with all of this, but I thought this
>was interesting reading. Cilkchess must have been strong.
>kburcham
>
>http://supertech.lcs.mit.edu/cilk/
>
>Cilkchess came in fourth in the 1999 world computer championships.
>Cilkchess beat Junior, Rebel, Nimzo, Chesstiger, Hiarcs, Diep. in the standings.

Actually i didn't play the big cilkchess but 'mini' in paderborn and
DIEP won from it. I did lose to cilkchess in the dutch champs though
where it participated.

Programmer is Don Dailey, perhaps he can react himself here. He's
a real cool guy. We played some game cilkchess - diep at my home.

In world champs i remember it usually searched around 17 ply.

Note that i outsearched any opponent in the endgame. I got 20 ply there.

Reason is my endgame eval was more stupid than theirs and i ran on
Bob's quad in wcc99.

Best regards,
Vincent

>But I did notice that Cilkchess did not have to play most of these in the
>tournament.
>Maybe luck of the draw, but still interesting.
>
>"Cilkchess is a parallel program which will be running on a 256-processor SGI
>Origin 2000
>at NASA Ames for the WCCC. Cilkchess won First Prize in the 1996 Dutch Open and
>took
>Second in both 1997 and 1998. Our earlier program, *Socrates, took Second in the
>1995 WCCC, tying the winner Fritz in the main part of the tournament, but losing
>in the playoff.
>
>Cilkchess is programmed in the Cilk multithreaded programming language, which
>allows
> highly irregular programs, such as chess, to be written with ease for parallel
>computers.
>The program uses a parallel variant of the MTD(f) search algorithm that
>incorporates
>null-move forward pruning, but few extensions. The evaluation function has been
>tuned
>from thousands of self-play games using a temporal-coherence learning algorithm.
>The
>transposition table is stored in 32 gigabytes of shared memory. In the late
>middle game,
>Cilkchess typically looks more than 15 ply (half-moves) ahead and performs 5-11
>million
>make-moves per second".
>
>http://www.uni-paderborn.de/~wccc99/
>
>[Event "WCCC99"]
>[Site "Paderborn"]
>[Date "1999.06.19"]
>[Round "7"]
>[White "Hiarcs"]
>[Black "Cilkchess"]
>[Result "0-1"]
>
>1. d4 e6 2. c4 c5 3. Nf3 Nf6 4. e3 cxd4 5. exd4 d5 6. Nc3 Nc6 7. Bg5 Be7 8. c5
>O-O
>9. Bb5 Bd7 10. O-O b6 11. Na4 bxc5 12. Bxf6 Bxf6 13. Nxc5 Be8 14. Rc1 Rc8
>15. Ba6 Rb8 16. b3 Qa5 17. a4 Rb4 18. Nd3 Rb6 19. Bb5 Rxb5 20. axb5 Nxd4
>21. Nxd4 Bxd4 22. Qc2 Bxb5 23. Rfd1 Qb6 24. Nc5 Be5 25. Qd2 Qb8 26. g3 Bf6
>27. Rc2 Qa8 28. f3 a5 29. Rdc1 Be7 30. Qd4 Bg5 31. f4 Bf6 32. Qd2 d4 33. Nd3 Be7
>34. Rd1 Qf3 35. Ne5 Qe4 36. Ra2 d3 37. Qg2 Qd4+ 38. Kh1 Bb4 39. Qb7 Qc5 40. Qe4
>d2
>41. Kg2 Qb6 42. Rc2 Rd8 43. g4 f6 44. Nc6 Qb7 45. Qxe6+ Kf8 46. Kf2 Re8 47. Qd5
>Bxc6
>48. Rxc6 Qa7+
>0-1



This page took 0 seconds to execute

Last modified: Thu, 15 Apr 21 08:11:13 -0700

Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.