Not logged inRybka Chess Community Forum
Up Topic Rybka Support & Discussion / Rybka Discussion / Rybka3 CCRL 40/40 results
- - By Uri Blass (*****) Date 2008-08-09 10:10
first result
Rybka3 64 bit 4 cpu-Zappa Mexico II 64-bit 4CPU 3073   22.5 − 9.5 (+14−1=17) (+17 -6 =20 for rybka2.3.2a)

CCRL rating

1)Rybka 3 64-bit 4CPU        3188
2)Rybka 2.3.2a 64-bit 4CPU 3129

link:
http://www.computerchess.org.uk/ccrl/4040.live/rating_list_all.html

More games are needed but the result suggest smaller improvement for rybka3 at longer time control and it seems that Zappa has good chances to win draw odd match against rybka at long time control.

I expect more than 80 elo improvement for rybka3 at 40/40 but I doubt if she will able to keep >100 elo improvement so I guess that at least 80 elo improvement by chessbase was right and there was no reason to claim at least 100 elo as some people suggested.

Uri
Parent - By Nelson Hernandez (Gold) Date 2008-08-09 10:31
I believe Larry has stated more than once that apparently Rybka 3 has the most difficulty against Zappa but trounces other engines.  Clearly we must wait for all the results to come in.
Parent - - By Venator (Silver) Date 2008-08-09 10:38
I wouldn't make any conclusion based on 32 games yet....
Parent - - By Uri Blass (*****) Date 2008-08-09 11:02
I also did not make conclusion.

The result suggest is not a definite conclusion and I expect more than 80 elo improvement inspite of the fact that the difference in rating is smaller.

I am of course not sure that the improvement is not going to be more than 100 elo but I think that chessbase was right when they did not promise at least 100 elo.

Uri
Parent - - By George Tsavdaris (****) Date 2008-08-09 11:38

>I am of course not sure that the improvement is not going to be more than 100 elo but I think that chessbase was right when they did not promise at least 100 elo.


On the other hand CEGT early results at blitz suggest a more than 150 ELO improvement, so perhaps Chessbase did wrong that did not say a +120 ELO improvement to promote Rybka 3 better and to attract more customers.
Actually it depends for what time control we speak about....
Parent - By Venator (Silver) Date 2008-08-09 12:01
I think it is easy for ChessBase to change the 80 number as soon as the official new CEGT and/or CCRL list comes out :-)
Parent - By Venator (Silver) Date 2008-08-09 12:02
I also did not make conclusion.

But you made a small rating list.

but I think that chessbase was right when they did not promise at least 100 elo.

We can only conclude this as soon as the official 40/20 and 40/120 rating lists come out. I understand ChessBase being conservative about the rating improvement, it can easily be changed as soon as official data are available.
Parent - By tralala (*) Date 2008-08-09 12:11
Interesting! This is the first result of some games on longer time controls (actually not very long, given it's calibrated towards an older Athlon Proc.). Well as others have pointed out we shouldn't read too much in such a small sample, however it would be no surprise if the improvement in longer time controls will be overproportionally smaller since improvements in tactics (and that seems the area where most of the improvements have been made) tend to be more visible in blitz.

Joachim
Parent - - By Geomusic (*****) Date 2008-08-09 13:07 Edited 2008-08-09 13:20
not enough games to decide anything yet of course. :)

But we all hope of course for the best :)

3250+ CCRL here we come!
Parent - By Ray (****) Date 2008-08-09 13:31
Too early to tell of course, but we have eight Quad Rybka games in progress from testers, so results will be known soon enough
We did get easily more than +100 ELO at Chess960 Blitz, and I expect more than +100 at 40/40 normal chess as well
Parent - - By lkaufman (*****) Date 2008-08-09 13:33
This result does show how hard it is to get a high rating at longer time controls; we won by 14 to 1 in decisive games, and it was a bad rating performance! As I've often said, the longer time controls mean more draws. Nevertheless, I cannot believe that if Rybka 3 holds a 140 Elo gain in blitz, that the rating gain at 40/40 could be under 100; that is too large a drop for this difference in time control, in my experience. I'll predict that when we play engines other than Zappa, we will be back into three digit gains.
Parent - - By tralala (*) Date 2008-08-09 14:56
Well such differences do exist. They are the exception but not impossible. If you compare the difference between Zappa Mexiko II and Rybka 2.3.2a on CCRL between 40/4 and 40/40 the difference is +35 Elo in favor for Zappa. That is no small difference isn't it?
And if you compare Zappa to Hiarcs the difference between the two time controls is +52 in favor of Zappa (Zappa being a poor blitzer and Hiarcs being a good one).
In Blitz it seems Zappa and Hiarcs are of comparable strength (within the error margins) but in longer time controls Zappa is clearly stronger. So it is possible that the performance of an engine is non-linear between different time controls.

The same picture holds true for CEGT, especially in the case Zappa-Hiarcs (less accentuated for Zappa-Rybka).
Parent - By lkaufman (*****) Date 2008-08-10 00:04
Yes, but since Rybka 3 is so much "smarter" than 2.3.2a, I would expect a relatively small drop (if any) with more time, and my own tests showed only a 2 Elo drop for a doubling of the time limit, though of course even with several hundred games this could be a substantial underestimate.
Parent - By Geomusic (*****) Date 2008-08-11 13:29
this is what expressed earlier we need more engines to play against erm...like crafty! LOL
Parent - - By turbojuice1122 (Gold) Date 2008-08-09 14:01
Hmmm...perhaps strange things happen when one does use the Elo rating formula...
Parent - By George Tsavdaris (****) Date 2008-08-09 14:35
And what formula he should use?
Parent - - By Uri Blass (*****) Date 2008-08-10 20:01
update of the ccrl 40/40 results

1)Rybka 3 64-bit 4CPU 3241 +56 −53 78.1% −185.5 38.8% 121   
2)Rybka 2.3.2a 64-bit 4CPU 3129 +15 −15 71.6% −143.4 41.1% 1500

New results of Rybka

Rybka3 64 bit 4CPU-Naum3.1 64 bit 4 cpu 41.5 − 10.5(+31−0=21)
Rybka3 64 bit 4CPU-Hiarcs 12 4cpu          30.5-6.5(+26 -2 =9)
 
Parent - - By Uri Blass (*****) Date 2008-08-11 17:18
Now Rybka3 has 3247 and new results are against Rybka2.3.2a toga and hiarcs.
Parent - By Uri Blass (*****) Date 2008-08-12 11:42
Now Rybka is leading with only 3233 after 227 games
still it is more than 100 elo better than the previous version

http://www.computerchess.org.uk/ccrl/4040.live/rating_list_all.html
Parent - - By gmnotyet (*) Date 2008-08-11 17:33
Wow, Naum 3.1 did not win a single game against Rybka 3 in 52(!) tries.

Rybka3 64 bit 4CPU-Naum3.1 64 bit 4 cpu 41.5 − 10.5(+31−0=21)
Parent - By Uly (Gold) Date 2008-08-11 19:09 Edited 2008-08-11 19:11

> Naum 3.1 did not win a single game against Rybka 3 in 52(!) tries.


He's the game Naum 3.1 won against Rybka 3 on my computer:

[Date "2008.08.09"]
[Round "1.4"]
[White "Rybka 3 1-cpu 32-bit"]
[Black "Naum 3.1"]
[Result "0-1"]
[Annotator "2. +0.39   2... +0.43"]
[PlyCount "100"]
[TimeControl "60+7"]

{Pondering=Off Start positions=e4nf.pgn} 1. e4 Nf6 2. e5 {0.39/13 13} Nd5 {
0.43/14 8} 3. Nc3 {0.39/12 5} e6 {0.39/14 13} 4. Nf3 {0.34/12 14} Nc6 {
0.39/14 11} 5. Nxd5 {0.30/12 15} exd5 {0.52/4 0} 6. d4 {0.30/12 5} d6 {
0.20/13 8} 7. c3 {0.35/11 9} dxe5 {0.38/12 13} 8. dxe5 {0.46/11 3} Bg4 {
0.36/13 16} 9. Bf4 {0.50/10 7} f6 {0.23/13 10} 10. Bb5 {0.49/10 8} Bc5 {
0.41/13 12} 11. exf6 {0.58/9 3} Qxf6 {0.35/13 13} 12. Qxd5 {0.14/10 14} Qxf4 {
0.43/13 5} 13. Bxc6+ {0.07/10 7} bxc6 {0.24/15 14} 14. Qxc6+ {0.07/10 12} Kf7 {
0.20/15 6} 15. Qxc5 {0.14/11 6} Rhe8+ {0.10/14 11} 16. Kf1 {0.14/9 0} Rab8 {
0.09/14 12} 17. Qd5+ {0.00/9 6} Kf6 {-0.35/13 6} 18. b3 {0.00/11 6} Rbd8 {
-0.28/13 7} 19. Qb7 {0.13/10 13} Qd6 {-2.67/14 5} 20. Nd4 {-2.02/9 11} c5 {
-2.23/15 10} 21. Qxa7 {-2.49/9 9} Ra8 {-2.18/14 7} 22. Qb7 {-2.49/9 9} cxd4 {
-2.41/13 7} 23. h3 {-3.28/9 8} Be2+ {-4.52/13 7} 24. Kg1 {-3.28/8 0} dxc3 {
-4.64/14 7} 25. g3 {-3.66/10 10} c2 {-4.64/12 6} 26. Kh2 {-3.91/10 8} Rab8 {
-5.20/14 6} 27. Qg2 {-4.84/10 6} Bd1 {-5.53/15 7} 28. Rf1 {-5.01/10 8} Qd4 {
-5.53/13 7} 29. Rc1 {-5.01/10 8} Qd2 {-5.53/13 6} 30. Ra1 {-5.15/10 8} c1=Q {
-5.73/13 5} 31. Rxc1 {-5.00/9 7} Qxc1 {-6.04/14 6} 32. b4 {-5.28/10 7} Qc2 {
-6.28/14 8} 33. Qd5 {-5.66/9 7} Rbd8 {-6.28/14 3} 34. Qb7 {-6.27/10 7} Qxa2 {
-6.43/10 6} 35. Qc7 {-5.76/8 7} Qd2 {-6.89/12 7} 36. Qc5 {-6.53/9 7} Rd6 {
-7.04/14 9} 37. Kg1 {-7.33/9 7} Bf3 {-7.09/13 7} 38. g4 {-7.37/10 7} Re5 {
-7.56/14 6} 39. Qc8 {-7.75/9 7} Bc6 {-9.16/14 6} 40. Qg8 {-8.89/8 7} Bb5 {
-9.86/14 9} 41. Qa8 {-11.14/8 7} Bxf1 {-13.09/13 7} 42. Kxf1 {-11.59/9 7} Kg6 {
-#9/7 0} 43. Kg2 {-#13/9 3} Rf6 {-#8/5 0} 44. Qa7 {-#13/7 0} Qd5+ {-#7/3 0} 45.
Kh2 {-#10/5 0} Re1 {-#6/2 0} 46. f4 {-#10/3 0} Rh1+ {-#5/1 0} 47. Kg3 {-#8/3 0}
Qd3+ {-#4/1 0} 48. Qe3 {-#7/3 0} Qxe3+ {-#3/1 0} 49. Kg2 {-#3/3 0} Rg1+ {
-#2/1 0} 50. Kh2 {-#2/3 0} Qg3# {-#1/1 0 Checkmate !!} 0-1

Naum also achieved a draw, and 13 loses (results too loopsided.)
Parent - By Martin Thoresen (***) Date 2008-08-12 15:53
Current score after 30 of 100 games for CCRL 40/40 between Rybka 3 64-bit 4CPU and Glaurung 2.1 64-bit 4CPU is +26 -0 =4.

I'm gonna run another match after this so I'll keep you updated.

Best Regards,
Martin
CCRL
Up Topic Rybka Support & Discussion / Rybka Discussion / Rybka3 CCRL 40/40 results

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill