Not logged inRybka Chess Community Forum
Up Topic Rybka Support & Discussion / Rybka Discussion / Rybka in Leiden
1 2 3 4 Previous Next  
Parent - By NATIONAL12 (Gold) Date 2008-06-01 21:58
i have never smoked a Gurkha but have had good reports of them. a word of warning however if anyone ever offers you a Guantanmera to smoke ram it straight down his throat it is the most disgusting cigar i have ever smoked.
Parent - By Lukas Cimiotti (Bronze) Date 2008-06-01 20:38
I guess my computer - working for the Rybka team - was the fastest in that tournament. Hiarcs team had a computer that was only 10% behind. So for the top teams there were no significant differences.
Btw. my comp didn't cost $20.000 - it's something around €6000 - but you can reduce that to €4000 without significant restrictions.
And it's possible to build an 8 cores comp at 2.8 GHz at a price of less than 2000€.

Regards,
Lukas
Parent - - By Kappatoo (*****) Date 2008-06-01 16:31
People keep asking for engines with a unique playing style, and then they still complain.
To Vas: Will Rybka 3 come with a Tjezzz personality?
Parent - - By Vasik Rajlich (Silver) Date 2008-06-01 17:45
That's probably for Rybka 5 :)

Vas
Parent - By Kappatoo (*****) Date 2008-06-01 18:21
By the time Rybka 5 is released, maybe Tjezzz will have a 'Rybka' personality.
Parent - - By Venator (Silver) Date 2008-06-01 15:40
Round 9 saw a long game against Joker. Rybka won, which means Rybka is the winner of the ICT Leiden tournament with a 8.5/9 score. Surprisingly of all participants Joker made the most theory moves against Rybka :-). Then a well known position remained, with white a pawn up and black having compensation. Rybka quickly grasped the initiative, but the ending was still difficult to win. 45.Rd1? was a huge blunder, after which the game was over.

Final standings:

1. Rybka 8.5/9
2. Deep Sjeng 8.0
3. Hiarcs 6.5
4. Diep 6.5

Also congrats to the other prize winners and especially to Gian Carlo for this excellent result with Deep Sjeng. Normally 8/9 would be good enough to win this tournament :-).

[Event "ICT Leiden"]
[Site "Apeldoorn"]
[Date "2008.06.01"]
[Round "9"]
[White "Joker"]
[Black "Rybka"]
[Result "0-1"]
[ECO "B80"]
[PlyCount "146"]
[TimeControl "4200"]

{1MB, Leiden1.ctg, Intel Core Duo T2300} 1. e4 {[%emt 0:00:00]} c5 {
[%emt 0:00:00]} 2. Nf3 {[%emt 0:00:08]} d6 {[%emt 0:00:00]} 3. d4 {
[%emt 0:00:06]} cxd4 {[%emt 0:00:00]} 4. Nxd4 {[%emt 0:00:05]} Nf6 {
[%emt 0:00:00]} 5. Nc3 {[%emt 0:00:06]} a6 {[%emt 0:00:00]} 6. Be3 {
[%emt 0:00:05]} e6 {[%emt 0:00:00]} 7. f3 {[%emt 0:00:07]} b5 {[%emt 0:00:00]}
8. Qd2 {[%emt 0:00:06]} Nbd7 {[%emt 0:00:00]} 9. g4 {[%emt 0:00:06]} b4 {
[%emt 0:00:00]} 10. Na4 {[%emt 0:00:06]} h6 {[%emt 0:00:00]} 11. O-O-O {
[%emt 0:00:06]} Ne5 {[%emt 0:00:00]} 12. Qxb4 {[%emt 0:00:08]} Bd7 {
[%emt 0:00:00]} 13. Bf4 {[%emt 0:00:07]} g5 {[%emt 0:00:00]} 14. Bd2 {
[%emt 0:00:06]} Be7 {[%emt 0:00:00]} 15. Qb3 {[%emt 0:01:26]} Rb8 {
[%emt 0:00:00]} 16. Ba5 {[%emt 0:01:36]} Rxb3 {[%emt 0:00:00]} 17. Bxd8 {
[%emt 0:02:40]} Rb8 {[%emt 0:00:00]} 18. Bxe7 {[%emt 0:02:00]} Kxe7 {
[%emt 0:00:00]} 19. Nc3 {[%emt 0:01:43]} h5 {[%emt 0:00:00]} 20. gxh5 {
[%emt 0:02:19]} a5 {[%emt 0:00:00]} 21. Rg1 {[%emt 0:02:39]} Rbg8 {
[%emt 0:00:00]} 22. a3 {[%emt 0:01:53]} Nxh5 {[%eval 26,20] [%emt 0:02:01]} 23.
b4 {(Rd2) [%emt 0:03:17]} axb4 {[%eval -17,20] [%emt 0:01:43]} 24. axb4 {
[%emt 0:01:25]} Nf4 {[%eval -19,21] [%emt 0:01:01]} 25. Rd2 {[%emt 0:02:56]}
Rc8 {[%eval -26,20] [%emt 0:00:19]} 26. Na2 {[%emt 0:03:10]} f6 {
[%eval -36,19] [%emt 0:00:16]} 27. Kb1 {[%emt 0:02:05]} d5 {
[%eval -41,19] [%emt 0:00:19]} 28. Kb2 {(Rg3) [%emt 0:03:14]} Rc4 {
[%eval -55,17] [%emt 0:00:41]} 29. Rg3 {(exd5) [%emt 0:01:24]} Ned3+ {
[%eval -79,16] [%emt 0:00:49]} 30. Bxd3 {[%emt 0:01:09]} Rxd4 {
[%eval -91,19] [%emt 0:00:16]} 31. exd5 {(Rg1) [%emt 0:02:23]} e5 {
[%eval -92,19] [%emt 0:02:14]} 32. Rg1 {[%emt 0:00:17]} Bb5 {
[%eval -89,19] [%emt 0:02:17]} 33. Ka3 {(Kb3) [%emt 0:02:18]} Rc8 {
[%eval -91,19] [%emt 0:02:04]} 34. Kb3 {[%emt 0:01:23]} Nxd3 {
[%eval -104,21] [%emt 0:01:21]} 35. cxd3 {[%emt 0:00:40]} Bxd3 {
[%eval -110,22] [%emt 0:00:13]} 36. Kb2 {[%emt 0:00:58]} Bc4 {
[%eval -90,22] [%emt 0:01:34]} 37. Rxd4 {(d6+) [%emt 0:01:47]} exd4 {
[%eval -112,20] [%emt 0:00:01]} 38. Rd1 {[%emt 0:00:47]} Rh8 {
[%eval -101,24] [%emt 0:00:26]} 39. d6+ {[%emt 0:00:42]} Kd7 {
[%eval -100,24] [%emt 0:00:01]} 40. Rd2 {[%emt 0:01:50]} Rh4 {
[%eval -109,26] [%emt 0:00:16]} 41. Nc1 {[%emt 0:00:57]} Kxd6 {
[%eval -114,26] [%emt 0:02:28]} 42. Nb3 {[%emt 0:01:13]} Kd5 {
[%eval -114,25] [%emt 0:00:18]} 43. Na5 {[%emt 0:01:00]} Ba6 {
[%eval -134,24] [%emt 0:02:27]} 44. Nb3 {(Rf2) [%emt 0:01:29]} Kc4 {
[%eval -103,22] [%emt 0:04:33]} 45. Rd1 {(Nc5) [%emt 0:00:50]} Rxh2+ {
[%eval -273,23] [%emt 0:01:39]} 46. Ka3 {[%emt 0:00:59]} d3 {
[%eval -293,23] [%emt 0:00:58]} 47. Nc5 {(Rc1+) [%emt 0:01:22]} Kc3 {
[%eval -293,21] [%emt 0:00:01]} 48. Nxa6 {[%emt 0:01:52]} d2 {
[%eval -298,16] [%emt 0:00:01]} 49. b5 {[%emt 0:01:26]} Kc2 {
[%eval -509,20] [%emt 0:01:00]} 50. Rf1 {(Ra1) [%emt 0:00:22]} d1=Q {
[%eval -509,20] [%emt 0:00:01]} 51. Rxd1 {(Nb4+) [%emt 0:00:45]} Kxd1 {
[%eval -441,20] [%emt 0:00:01]} 52. b6 {(Nc5) [%emt 0:01:29]} Rh7 {
[%eval -441,19] [%emt 0:00:01]} 53. Nc5 {(Nc7) [%emt 0:01:49]} Ke2 {
[%eval -608,13] [%emt 0:00:36]} 54. Kb4 {(b7) [%emt 0:00:24]} Kxf3 {
[%eval -496,16] [%emt 0:00:01]} 55. b7 {(Nd3) [%emt 0:02:33]} Rxb7+ {
[%eval -1481,14] [%emt 0:02:19]} 56. Nxb7 {(Ka4) [%emt 0:01:41]} g4 {
[%eval 1,3] [%emt 0:00:01]} 57. Nc5 {(Nd6) [%emt 0:00:45]} g3 {
[%eval -32751,3] [%emt 0:00:01]} 58. Nd7 {(Nd3) [%emt 0:01:25]} g2 {
[%eval -32752,3] [%emt 0:00:01]} 59. Nxf6 {[%emt 0:00:59]} g1=Q {
[%eval -32753,3] [%emt 0:00:00]} 60. Kc4 {[%emt 0:00:35]} Kf4 {
[%eval -32754,3] [%emt 0:00:00]} 61. Nd7 {[%emt 0:00:30]} Qg4 {
[%eval -32755,3] [%emt 0:00:00]} 62. Nb6 {(Nf6) [%emt 0:00:46]} Ke5+ {
[%eval -32756,3] [%emt 0:00:01]} 63. Kc5 {[%emt 0:00:59]} Qd4+ {
[%eval -32757,3] [%emt 0:00:00]} 64. Kb5 {(Kc6) [%emt 0:00:32]} Qc3 {
[%eval -32758,3] [%emt 0:00:01]} 65. Nc4+ {(Ka4) [%emt 0:00:23]} Kd4 {
[%eval -32759,3] [%emt 0:00:01]} 66. Nb6 {[%emt 0:00:09]} Qc5+ {
[%eval -32760,3] [%emt 0:00:00]} 67. Ka6 {[%emt 0:00:09]} Qb4 {
[%eval -32761,3] [%emt 0:00:00]} 68. Ka7 {[%emt 0:00:09]} Kc5 {
[%eval -32762,3] [%emt 0:00:00]} 69. Nd7+ {[%emt 0:00:10]} Kd6 {
[%eval -32763,3] [%emt 0:00:00]} 70. Nb8 {[%emt 0:00:09]} Qb5 {
[%eval -32764,3] [%emt 0:00:00]} 71. Na6 {(Ka8) [%emt 0:00:09]} Kd7 {
[%eval -32765,3] [%emt 0:00:01]} 72. Nc5+ {(Ka8) [%emt 0:00:09]} Kc8 {
[%eval -32766,3] [%emt 0:00:01]} 73. Na4 {(Ka8) [%emt 0:00:09]} Qa5# {
[%eval -32767,1] [%emt 0:00:01]} 0-1
Parent - By Laszlo (***) Date 2008-06-01 16:27
Bravo, bravo for the last Najdorf and the final victory!
Parent - - By sarciness (***) Date 2008-06-01 16:28
This Rybka is good...

Is it really only using 1Mb hash?
Parent - By Lukas Cimiotti (Bronze) Date 2008-06-01 17:04
No - 2GB
Parent - - By turbojuice1122 (Gold) Date 2008-06-01 17:30
Congratulations on the win and the impressive score!
Parent - - By Venator (Silver) Date 2008-06-01 18:47
Thanks! It was Rybka's 5th tournament in Leiden if I counted correctly and Rybka won them all. She scored 8.5, 9.0, 7.5, 8.0 and 8.5 points in 9 games respectively, not bad I would say :-).
Parent - - By George Tsavdaris (****) Date 2008-06-01 20:57 Edited 2008-06-01 21:01
If we have :
Rybka 3.0 on this octal hardware with the private book of yours it played in Leiden, against
an identical Rybka 3.0 on an identical octal hardware, but with the commercial Rybka3.ctg book it will have with the final Rybka 3 package,
then:
How many ELO Rybka 3.0 with your private book, will have more than Rybka 3.0 with the publicly available Rybka3 book?

(Obviously i ask for an estimation. A rough estimation if it's not possible a more concrete one.)

And what about Rybka 3 with your private book on a single CPU against the same R3 on a single CPU but with Rybka3 commercial book?
I want this estimation too, because perhaps on octals the importance of book becomes lesser as strength increases(diminishing returns).
Parent - - By Venator (Silver) Date 2008-06-02 05:44
Because of the setup of the private book it is quite pointless to test this. The Rybka3.ctg book will be very wide, with lots of variety and many lines to be played. This could be a disadvantage, but in matches with many games this might turn out to be an advantage. In Leiden I played with both books - private and Rybka3.ctg.
Parent - By Bouddha (****) Date 2008-06-02 13:09
which games were played with Rybka3.ctg ?
Parent - - By turbojuice1122 (Gold) Date 2008-06-01 17:35
By the way, would you say that Rybka is a fairly strong chess engine? ;-)
Parent - By Venator (Silver) Date 2008-06-01 18:48
I would skip the 'fairly' and replace it by 'very' :-)
Parent - - By Sesse (****) Date 2008-06-01 18:50
Congratulations; another fine win! There's not much of a margin here, though, as you point out -- I guess that's how it is in any really competitive field. :-)

/* Steinar */
Parent - By Venator (Silver) Date 2008-06-01 18:54
Thanks! Well, if the 2nd place program wins all its games except against the no. 1, the maximum margin you can get is 1 point. Deep Sjeng really played a great tournament here, normally 8/9 would clinch it quite easily.
Parent - - By Nick (*****) Date 2008-06-01 19:15
That's pretty impressive theory H-G .. did you spend much time on your book? 
Parent - - By Venator (Silver) Date 2008-06-01 20:27
Harm Geert explained he got a PGN book by another chess programmer, who wanted him to try this out. After black's Be7 Joker was out of book, but followed theory another couple of moves, reaching the well known position after move 21. These moves are not hard to find, but if anybody would tell me before the tournament that Joker-Rybka would be the most interesting game from a theoretical point of view, I would not believe him :-).

Anyway, Joker played a decent game and if it had not blundered by Rd1? I am still not sure if the position is won for black.
Parent - - By ernest (****) Date 2008-06-02 16:37
> ... and if it had not blundered by Rd1?

Well, Rybka 2.3.2a mp herself on a 3GHz Core 2 Duo doesn't avoid Rd1 so fast...!
Here are several infinite analysis (of course non-identical) tries.

8/8/b4p2/6p1/1Pkp3r/1N3P2/1K1R3P/8 w - -


Engine: Rybka 2.3.2a mp 32-bit (512 MB) by Vasik Rajlich
14.01   0:01   -1.18   45.Rd1 Bb5 46.Ka3 Kd5 47.Rc1 Be2 48.Rc5+ Ke6 49.Rc6+ Ke7 50.Rc2 d3 51.Rd2 (258.644) 260
15.01   0:02   -0.98   45.Rd1 Bb5 46.Ka3 Kd5 47.Rc1 Kd6 48.Rd1 d3 49.Nc5 Rh3 50.Kb3 Rxf3 (611.618) 270
16.01   0:03   -0.97   45.Rd1 Bb5 46.Ka3 Kd5 47.Rc1 Kd6 48.Rd1 Ke5 49.Re1+ Kf5 50.Rc1 Be2 51.b5 Bxf3 (892.313) 260
17.01   0:06   -1.07   45.Rd1 Rxh2+ 46.Ka3 d3 47.Nc5 Bb5 48.Rxd3 Rh3 49.Rb3 Bc6 50.b5 Kxc5 51.bxc6 Kxc6 (1.668.653) 256
18.01   0:09   -1.06   45.Rd1 Rxh2+ 46.Ka3 d3 47.Nc5 Bb5 48.Rxd3 Rh3 49.Rb3 Bc6 50.b5 Kxc5 51.bxc6 Kxc6 (2.299.226) 256
19.01   0:12   -1.87   45.Rd1 Rxh2+ 46.Ka3 d3 47.Nc5 d2 48.Nxa6 Kc3 49.b5 Rh8 50.Rb1 Kc2 51.Rb2+ Kc1 (3.347.796) 264
19.02   0:20   -1.17   45.Nc5 Bb5 46.Ne4 Rf4 47.Kc1 Bd7 48.Rb2 Bf5 49.Nd6+ Kd5 50.Nxf5 Rxf5 51.b5 Rxf3 (5.256.666) 257  TB:15
20.01   0:33   -1.22   45.Nc5 Bb5 46.Ne4 Rf4 47.Kc1 Rxf3 48.Nd6+ Kxb4 49.Rxd4+ Kc5 50.Nxb5 f5 51.h4 g4 (8.728.572) 263  TB:24
best move: Nb3-c5 time: 0:41.641 min  n/s: 269.901  nodes: 10.926.255 TB: 42
15.01   0:01   -0.94   45.Rd1 Bb5 46.Ka3 Kd5 47.Rc1 Bc4 48.Nd2 d3 49.Ka4 Ba6 50.Ne4 Rxh2 (500.043) 262
16.01   0:02   -1.09   45.Rd1 Bb5 46.Ka3 Kd5 47.Rc1 Bc4 48.Nd2 d3 49.h3 Ba6 50.Rc5+ Kd4 51.Rc6 Bb7 (676.554) 267
17.01   0:06   -0.97   45.Rd1 Rxh2+ 46.Ka3 Kc3 47.Rc1+ Kd3 48.Nc5+ Kd2 49.Nb3+ Ke3 50.Rc6 Bb5 51.Rxf6 d3 (1.736.360) 265
18.01   0:10   -0.90   45.Rd1 Rxh2+ 46.Ka3 Kc3 47.Rc1+ Rc2 48.Rxc2+ Kxc2 49.Nxd4+ Kd3 50.Ne6 Bc4 51.Nc7 f5 (2.666.501) 251  TB:7
19.01   0:19   -0.98   45.Rd1 Rxh2+ 46.Ka3 Kc3 47.Rc1+ Rc2 48.Rxc2+ Kxc2 49.Nxd4+ Kd3 50.Nf5 Bb7 51.b5 Bxf3 (4.899.045) 256  TB:70
20.01   0:24   -1.15   45.Rd1 Rxh2+ 46.Ka3 Kc3 47.Rc1+ Rc2 48.Rxc2+ Kxc2 49.Nxd4+ Kd3 50.Nf5 Bb7 51.b5 Bxf3 (6.156.152) 257  TB:80
21.01   0:34   -1.28   45.Rd1 Rxh2+ 46.Ka3 Kc3 47.Rc1+ Rc2 48.Rxc2+ Kxc2 49.Nxd4+ Kd3 50.Nf5 Bb7 51.b5 Bxf3 (8.954.833) 263  TB:92
22.01   1:15   -1.96   45.Rd1 Rxh2+ 46.Ka3 d3 47.Nc5 Kc3 48.Nxa6 d2 49.b5 Kc2 50.Ra1 d1Q 51.Rxd1 Kxd1 (19.769.704) 268  TB:159
22.03   2:05   -1.13   45.Nc5 Bb5 46.Ne4 Rf4 47.Kc1 Bd7 48.Rb2 Bf5 49.Nd6+ Kd5 50.Nxf5 Rxf5 51.b5 Rxf3 (31.840.144) 260  TB:169
best move: Nb3-c5 time: 2:10.593 min  n/s: 261.950  nodes: 33.407.444 TB: 172
15.01   0:01   -1.02   45.Nc5 Bb5 46.Ne4 Rf4 47.Kc1 Rxf3 48.Nd6+ Kxb4 49.Rxd4+ Kc5 50.Nxb5 Kxb5 51.h4 Kc5 (448.346) 290
16.01   0:02   -1.18   45.Nc5 Bb5 46.Ne4 Rf4 47.Kc1 Bd7 48.Rb2 Bf5 49.Nd6+ Kd5 50.Nxf5 Rxf5 51.Rf2 Kc4 (746.002) 292
16.02   0:05   -0.97   45.Rd1 Bb5 46.Ka3 Kd5 47.Rc1 Kd6 48.Rd1 Ke5 49.Re1+ Kf5 50.Rc1 Kg6 51.Rc2 d3 (1.292.245) 258
17.01   0:08   -1.04   45.Rd1 Rxh2+ 46.Ka3 d3 47.Nc5 Bb5 48.Rxd3 Rh3 49.Rb3 Bc6 50.b5 Kxc5 51.bxc6 Kxc6 (2.016.540) 252
18.01   0:10   -1.04   45.Rd1 Rxh2+ 46.Ka3 d3 47.Nc5 Bb5 48.Rxd3 Rh3 49.Rb3 Bc6 50.b5 Kxc5 51.bxc6 Kxc6 (2.523.813) 252
19.01   0:13   -1.07   45.Rd1 Rxh2+ 46.Ka3 d3 47.Nc5 Bb5 48.Rxd3 Rh3 49.Rb3 Bc6 50.b5 Kxc5 51.bxc6 Kxc6 (3.472.984) 256
20.01   0:22   -1.30   45.Rd1 Rxh2+ 46.Ka3 d3 47.Nc5 Bb5 48.Rxd3 Rh3 49.Rb3 Bc6 50.b5 Kxc5 51.bxc6 Kxc6 (5.816.040) 259  TB:7
21.01   1:03   -2.17   45.Rd1 Rxh2+ 46.Ka3 d3 47.Nc5 Bb5 48.Rxd3 Rh3 49.Rb3 Bc6 50.b5 Kxc5 51.bxc6 Kxc6 (16.072.726) 258  TB:25
21.02   1:22   -1.18   45.Nc5 Bb5 46.Kc1 Rf4 47.Ne6 Rxf3 48.Nxd4 Rf1+ 49.Kc2 Kxb4 50.Nxb5 Kxb5 51.Kd3 Kc5 (20.475.065) 255  TB:26
22.01   1:34   -1.18   45.Nc5 Bb5 46.Kc1 Rf4 47.Ne6 Rxf3 48.Nxd4 Rf1+ 49.Kc2 Kxb4 50.Nxb5 Kxb5 51.Kd3 Kc5 (23.468.772) 253  TB:35
best move: Nb3-c5 time: 1:44.719 min  n/s: 257.650  nodes: 26.348.748 TB: 39
Parent - - By Venator (Silver) Date 2008-06-02 17:44
Interesting. Rybka in Leiden never considered 45.Rd1 and immediately made a big score jump as soon as Joker played that move. Probably the TB's already did the job there, or new Rybka is much better in this position.
Parent - - By Felix Kling (Gold) Date 2008-06-02 18:14 Edited 2008-06-02 18:21
Rybka seems to be simply better : 

[-1.35]  d=14  1.Nc5 Bb5 2.Ne4 Rf4 3.Kc1 Bd7 4.Rb2 Bf5 5.Nxf6 Rxf3 6.Rd2 Rb3 7.Ra2 Kxb4 (0:00.23)
[-1.33]  d=13  1.Nc5 Bb5 2.Ne4 Rf4 3.Kc1 Bd7 4.Rb2 Bf5 5.Nd6 Kd5 6.Nxf5 Rxf5 7.Rf2 Kc4 (0:00.12)
[-1.33]  d=12  1.Nc5 Bb5 2.Ne4 Rf4 3.Kc1 Bd7 4.Rb2 Bf5 5.Nd6 Kd5 6.Nxf5 Rxf5 7.Rf2 Kc4 (0:00.07)
[³/-0.97]  d=11  1.Nc5 Bb5 2.Ne4 Rf4 3.Kc1 Rxf3 4.Nd6 Kxb4 5.Rxd4 Kc5 6.Nxb5 Kxb5 7.h4 Kc5 (0:00.03)
[-1.04]  d=10  1.Nc5 Bb5 2.Ne4 Rf4 3.Kc1 Rxf3 4.Nd6 Kxb4 5.Rxd4 Kc5 6.Nxb5 Kxb5 7.h4 Kc5 (0:00.02)
[-1.08]  d=9  1.Nc5 Bb5 2.Ne4 Rf4 3.Kc1 Rxf3 4.Nd6 Kxb4 5.Rxd4 Kc5 6.Nxb5 Kxb5 7.h4 (0:00.02)
[-1.22]  d=8  1.Nc5 Bb5 2.Ne4 Rf4 3.Kc1 Rxf3 4.Nd6 Kxb4 5.Rxd4 Kc5 (0:00.01)
[-1.33]  d=8  1.Nc5 Bb5 2.Ne4 Rf4 (0:00.01)
[³/-0.93]  d=7  1.Nc5 Bb5 2.Ne4 Rf4 (0:00.00)
[³/-0.89]  d=6  1.Nc5 Bb5 2.Ne4 (0:00.00)
[-0.66]  d=5  1.Nc5 Bb5 (0:00.00)
[-0.63]  d=4  1.Nc5 (0:00.00)
[-0.61]  d=3  1.Nc5 (0:00.00)
[-0.59]  d=2  1.Nc5 (0:00.00)

(Rybka 2.3.2e13, not the latest version...)

However, it takes still some time when you enter Rd1 till the score is really dropping:
[µ/-1.77]  d=19  1...Rxh2 2.Ka3 d3 3.Nc5 d2 4.Nxa6 Kc3 5.b5 Kc2 6.Rf1 d1Q 7.Nb4 Kd2 8.Rxd1 (0:02.02)
[-1.37]  d=18  1...Rxh2 2.Ka3 d3 3.Nc5 d2 4.Nxa6 Kc3 5.b5 Kc2 6.Rf1 d1Q 7.Nb4 Kd2 8.Rxd1 (0:01.10)
[-1.37]  d=17  1...Rxh2 2.Ka3 d3 3.Nc5 d2 4.Nxa6 Kc3 5.b5 Kc2 6.Rf1 d1Q 7.Nb4 Kd2 8.Rxd1 (0:00.57)
[-1.39]  d=16  1...Rxh2 2.Ka3 d3 3.Nc5 d2 4.Nxa6 Kc3 5.b5 Kc2 6.Rf1 d1Q 7.Nb4 Kd2 8.Rxd1 (0:00.30)
[-1.37]  d=15  1...Rxh2 2.Ka3 d3 3.Nc5 d2 4.Nxa6 Kc3 5.b5 Kc2 6.Rf1 d1Q 7.Nb4 Kd2 8.Rxd1 (0:00.15)
[-1.36]  d=14  1...Rxh2 2.Ka3 d3 3.Nc5 d2 4.Nxa6 Kc3 5.b5 Kc2 6.Rf1 d1Q 7.Nb4 Kd2 8.Rxd1 (0:00.07)
[-1.36]  d=13  1...Rxh2 2.Ka3 d3 3.Nc5 d2 4.Nxa6 Kc3 5.b5 Kc2 6.Rf1 d1Q 7.Rxd1 Kxd1 8.b6 (0:00.05)
[-1.56]  d=13  1...Rxh2 2.Ka3 d3 3.Nc5 d2 4.Nxa6 Kc3 5.b5 Kc2 6.Rf1 d1Q 7.Rxd1 Kxd1 (0:00.04)
[-1.16]  d=10  1...Rxh2 2.Ka3 d3 3.Nc5 d2 4.Nxa6 Kc3 5.b5 Kc2 6.Ra1 d1Q 7.Rxd1 Kxd1 8.b6 (0:00.01)
[-1.34]  d=9  1...Rxh2 2.Ka3 d3 3.Nc5 d2 4.Nxa6 Kc3 5.b5 Kc2 6.Ra1 d1Q 7.Rxd1 Kxd1 8.b6 (0:00.00)
[-1.32]  d=7  1...Rxh2 2.Ka3 d3 3.Nc5 d2 4.Nxa6 Kc3 5.b5 Kc2 (0:00.00)
[-1.39]  d=5  1...Rxh2 2.Ka3 d3 3.Nc5 d2 4.Nxa6 Kc3 5.b5 Kc2 (0:00.00)
[-1.32]  d=4  1...Rxh2 2.Ka3 d3 3.Nc5 d2 4.Nxa6 Kc3 5.b5 Kc2 (0:00.00)
[-1.39]  d=2  1...Rxh2 2.Ka3 d3 3.Nc5 d2 4.Nxa6 Kc3 5.b5 Kc2 (0:00.00)

Of course on Lukas' monster hardware it dropped much faster I guess :)
Parent - - By Vasik Rajlich (Silver) Date 2008-06-02 21:13

> Of course on Lukas' monster hardware it dropped much faster I guess


Not to mention 64-bit OS :)

Vas
Parent - - By Felix Kling (Gold) Date 2008-06-02 21:37
What do you want to tell me? ;)

Where is the native Linux version? Where is Aquarium GUI for Linux? When both are ready, I'll buy a new computer and install a 64 bit OS :)
Parent - - By Vasik Rajlich (Silver) Date 2008-06-02 21:53
Is that a promise? I'll see what Victor thinks ... :)

Vas
Parent - - By Felix Kling (Gold) Date 2008-06-02 22:15
Yeah, this is a promise - In the meantime I'll try to convince enough people to use Linux to create a user base :)
Parent - - By Nick (*****) Date 2008-06-02 22:23
I use Linux occasionally, I even tried installing Rybka into microwine (I wanted 64-bit obviously, but also the boost from the kernel large page support) - but I couldn't get netchess to work with Rybka when it was running in microwine, no moves came from the rybka+microwine combination back to netchess.  I managed to use the native Linux zappa using netchess on Linux without problems.

My idea was to use a CB gui connected to playchess.com on Windows on some other machine, but powered by a 64-bit Rybka running on Linux with large page support - I couldn't get it to work though.
Parent - - By Banned for Life (Gold) Date 2008-06-03 00:24
You can get large page support for any chess engine but Rybka :-) by running arena under 7-max. This produces an ~10% increase in node counts.
Parent - - By Nick (*****) Date 2008-06-03 08:12

> running arena under 7-max


Yes I recall you'd posted about 7-max ( http://www.7-max.com/ ) before, so I downloaded it but my experience matched yours, I couldn't get it to work with Rybka.

In February IBM released large (1mb in addition to the default 4k) page support on high end mainframes and it's a definite help for long-running memory-intensive (large working set) applications.  Sesse reports up to a 20% speed up due to huge page usage by Rybka, from the microwine readme:

On my machine, this has been benchmarked to as much as almost 20% faster in single-CPU tests with 1GB hash, but your mileage might vary.

It's all about wringing that last few % out of your kit, if you're at your overclocking ceiling then exploiting better memory management techniques is one way forwards.  Imagine turning Lukas' machine from a 4ghz Skulltrail to an effective 4.5+ghz Skulltrail without overclocking :)  Simplistic numbers - I'm sure it wouldn't be that good, also I don't know if Steinar's numbers compared Rybka running in Linux to Rybka running on Windows, there might be some other effect slowing it down even - I was interested to test it anyway.
Parent - By Banned for Life (Gold) Date 2008-06-03 11:41
Vas cleared up the mystery by revealing that he uses a higher level shared memory call which is not intercepted by 7-max (reference the original thread for details).

My experience was that my node counts under 7-max and arena would be 10-15% faster than under arena alone. Its always good to give a conservative estimate in these cases. My improvement might have been lower than Sesse because I was using a larger hash (up to 3.242 GB for Zappa). The larger hash means that even the large pages require frequent replacement in the TLB.

I saw comparable improvements when running under 7-max and CB GUI8 (GUI9 won't work because it hangs on the Turk).

Finally, the X% improvement in node count should actually be greater than the corresponding increase in processor clock speed. Getting the same improvement by overclocking would be expected when all relevant clock speeds are increased by X%.

Regards,
Alan
Parent - - By Felix Kling (Gold) Date 2008-06-03 07:24
A simple solution could be Wine support by Aquarium and a build in microwine that allows you to run the 64 bit version also.
Parent - - By Nick (*****) Date 2008-06-03 08:13
Does Aquarium connect to Playchess.com?  I thought that CB might have encrypted their connection protocol. 
Parent - - By Harvey Williamson (*****) Date 2008-06-03 08:42 Edited 2008-06-03 08:45
No It wont connect to playchess
Parent - By Nick (*****) Date 2008-06-03 16:39
ok thanks
Parent - - By Felix Kling (Gold) Date 2008-06-03 16:02
Indeed this is not possible, but Convekta could provide the same on their server. No need for playchess...
Parent - - By Nick (*****) Date 2008-06-03 16:40
The comp chess engine freaks will go wherever they can get the most games I would imagine - at least they'll be able to use their ctg books on the Convekta server once Aquarium is out (I hope).  And if there is a "no kibbing" facility for private games you might even see Alan play there also :)
Parent - - By Harvey Williamson (*****) Date 2008-06-03 16:46
Yes they can use their ctg books without learning and hope that only the green moves are played.
Parent - - By Nick (*****) Date 2008-06-03 16:53

> and hope that only the green moves are played


Yes the "playing red moves in a ctg" bug hits often, it even hit Bjorn in the Freestyle 6 finals.   This is one of the reasons I moved to Polyglot books for engine v engine play, there are no red moves in Polyglot books.
Parent - - By Banned for Life (Gold) Date 2008-06-03 17:15
there are no red moves in Polyglot books

What do you mean by this (I am totally ignorant on Polyglot books)? Ideally I guess there should be a function with playing weights for the different candidate moves and the ability of the user to input a function that allows everything from choosing randomly among the moves, to selecting only the very best move. Of course it should be possible for the user to override the weight function derived from number of games played, win percentage, and recent history. Does Polyglot allow something like this?

I've had the same problem that Bjorn had with the GUI selecting random moves from the book. I reinstalled the GUI and it went away, but not until after a game where I watched my engine open 1.f3 followed by 2.Kf2. I can't tell you how much of a handicap this is, but I can assure you that it is very significant.

Regards,
Alan
Parent - - By Nick (*****) Date 2008-06-03 17:46 Edited 2008-06-03 17:52

> What do you mean by this


In a ctg book the "red and green move concept" might be quite useful for humans looking at the book but not for an engine.  If the move is red in a ctg book (probability of being played = 0) then it does not appear at all at that point in the tree of a Polyglot book.  This is one of the reasons why a Polyglot book is 1/100th the size of a ctg book for the same playable moves.

> Does Polyglot allow something like this?


Yes, you can "add priority analysis" through the Polyglot merge-book function, this is how I get the best lines into my Polyglot book.   Also using the merge-book function you can import games to your book (merge-book is asymmetric, it's quite a beautifully efficient design, an overloaded "constructor" :)).  With Polyglot you can also make "uniform" books where all move probabilities at a point in the tree are the same (again this is equivalent to "add priority analysis" to an empty ctg in the CB gui with a pgn file containing many games, all either 1-0 or 0-1 of course).  You can also manually edit the probabilities of move play using scid, but I don't do that, life being short as it is.

> I can't tell you how much of a handicap this is, but I can assure you that it is very significant.


Yes I can believe it - I recall Bjorn saying that he wasted hours removing all the red moves from his ctg book for Freestyle 6 round n (n >1).  I've heard many grumbles from Playchess regulars along the same lines.

Incidentally, with Polyglot books you can easily play on Playchess.com, Marc Lacrosse (who is the world leader on Polyglot books having worked with Fabien on the spec) wrote an article of how to achieve this - see http://64.68.157.89/forum/viewtopic.php?t=17421&highlight= .  You can also play on ICC with the same book.  When I read this article from Marc my book making life for engine v engine became a lot simpler.
Parent - - By Banned for Life (Gold) Date 2008-06-03 18:20
Thanks Nick. Great stuff. I'll be reading up on this. My major concern is that I use my opening book rather than a full functioned database as a repository for positions and outcomes. I'm not sure this will be supported by Polyglot. Another concern is that I keep my novelties in the book by coloring the move red but giving it two exclamation points (hope springs eternal you know :-)). Is there a way to keep your best moves in a Polyglot book but not play them? I've always wished with .ctg books that they had more than two settings (tournament book or not tournament book) to make this easier.

Regards,
Alan
Parent - - By Nick (*****) Date 2008-06-03 18:45 Edited 2008-06-03 18:47

> Is there a way to keep your best moves in a Polyglot book but not play them?


Not that I know of ... perhaps Marc will weigh in on that question. I keep all my own work in PGNs, generally you can use PGNs as a base to build any format book you wish.  I didn't want to be locked into ctg, I mean, you wouldn't want CB to have you by the short and curlies would you? :)

EDIT: http://idioms.thefreedictionary.com/have+by+the+short+and+curlies
Parent - By Banned for Life (Gold) Date 2008-06-03 20:20
Its way too late for that. Chessbase has had me by the short and curlies for several years now.

Chessbase was a good place when I started doing the engine chess thing, but its really gone into the sewer over the last three years. Once the Rybka GUI (I can't force myself to use that gay name they came up with) comes out with its .ctg converter, I will explore moving to another server that does not give kibitzers with an enlarged proboscus immediate access to my games.

Regards,
Alan
Parent - - By turbojuice1122 (Gold) Date 2008-06-03 19:31
Bah, red moves, green moves; according to many books, I played a red move on the second move of a recent correspondence game...
Parent - - By Banned for Life (Gold) Date 2008-06-03 20:21
My favorite first move with white is generally given a ?! in most books.
Parent - By Venator (Silver) Date 2008-06-03 20:28
But not in RybkaII.ctg :-)
Parent - - By Felix Kling (Gold) Date 2008-06-03 22:09
guess why... :)
Parent - By Banned for Life (Gold) Date 2008-06-03 23:24
Because it has the highest win percentage? :-)
Parent - - By Venator (Silver) Date 2008-06-03 20:29
Let me guess: 1.e4 c5 2.Nf3 b6!?

BTW, red doesn't necessarely mean 'bad'. In some books 1.e4 e6 is red, but fully playable of course.
Up Topic Rybka Support & Discussion / Rybka Discussion / Rybka in Leiden
1 2 3 4 Previous Next  

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill