Not logged inRybka Chess Community Forum
Up Topic Rybka Support & Discussion / Rybka Discussion / New improvements in Rybka 3
1 2 Previous Next  
- - By premraj_n (**) Date 2007-12-17 17:22 Edited 2007-12-17 17:35
HI Vas,
I was not reading much in here for few months after Rybka 2.3.2a release and i think it would be good if you can tell us what are the new improvements that you are looking to implement in version 3 and looking at them we may suggest anything that is missing seriously in 2.3.2a or 2.3.2 that you may have to look in.

One thing i could say is Rybka (compared to most other Top engines) reaches depth very quick and it does miss out on better moves that are not analysed more deeply.

Thx,
Prem
Parent - - By Uly (Gold) Date 2007-12-17 17:50

> One thing i could say is Rybka (compared to most other Top engines) reaches depth very quick and it does miss out on better moves that are not analysed more deeply.


But probably this is what makes Rybka the strongest engine available. The other engines "waste" too much time on searching for this "better move", even though maybe it doesn't exist. Rybka on the other hand, gets sure that the move she's playing is sound enough, and therefore, most of the time she'll play the most solid move.

I don't like this super solid playing style, but probably it's the best possible style for chess (Otherwise, you'd expect the strongest entity on the world to play differently.)
Parent - - By Carl Bicknell (*****) Date 2007-12-17 17:53
Is that how you would describe Rybka's style of play: Super Solid?
Parent - By Uly (Gold) Date 2007-12-17 18:09
Yes, unless the opponent has a weakness, then Rybka knows how to exploit this weakness and win the game. So, the only way to draw against Rybka is to be super solid yourself, and the only way to win remains unknown (Unless you exploit some bug or run into positions that Rybka doesn't know how to play, but this isn't an easy task most of the time.)

Rybka's style seems to be based on the logic that the enemy is going to make a mistake eventually, all you have to do is wait for this mistake to happen and then just use this mistake to your advantage and win. What do you do while you are waiting? To be solid and avoid to make a mistake yourself. This style seems to work.
Parent - - By turbojuice1122 (Gold) Date 2007-12-17 20:21
The strongest chess entity in the world would be Zappa Mexico on one of Suj's machines, and Zappa certainly has a very different playing style and technique.  As Suj says, it's one problem is that it searches too hard for opponent refutations, and the result is often a draw.
Parent - - By BB (****) Date 2007-12-17 20:26
The strongest chess entity in the world would be Zappa Mexico on one of Suj's machines

Are centaurs non-entities?
Parent - By turbojuice1122 (Gold) Date 2007-12-17 20:44
Centaurs are two or more entities, so this doesn't count :-).
Parent - - By Banned for Life (Gold) Date 2007-12-17 20:57
it's one problem is that it searches too hard for opponent refutations, and the result is often a draw.

This is nonsense. It was originally a theory from AC to explain why Zappa didn't do better when it was playing on a 512 processor machine. If you use Zappa for long term analysis, it is readily apparent that it just ain't so. As with other good engines, higher time-processing speed products generally produce more accurate evaluations, but not evaluations that are closer to zero.

Regards,
Alan
Parent - - By turbojuice1122 (Gold) Date 2007-12-17 21:02
I don't know about a theory from Anthony, but it's Suj's direct observation, as he has done many tests with that on 32 and 64 cores.  Even so, it trounces everything in sight, including Rybka, unless the newest version has great improvement from 4 to 8 to 16 to 32 processors.
Parent - - By Banned for Life (Gold) Date 2007-12-17 21:38
If you try to substantiate this claim with real game data, you will see it is a myth. Feel free to post a link to a large quantity of games showing that using Zappa on a 32 or 64 processor machine will lead to a statistically significant increase in the percentage of draws.

Alan
Parent - - By turbojuice1122 (Gold) Date 2007-12-17 22:26
As far as I know, we don't have any game data, so we just have to go by Suj's word, unless he's willing to post it himself.  I'm pretty sure that he's not part of the Zappa team.
Parent - - By Banned for Life (Gold) Date 2007-12-17 22:42
Suj is great guy, but I wouldn't rely on casual observations from him, or anyone else, to establish the asymptotic behavior of chess engines in general. Furthermore, I'm surprised that anyone who considers themselves to be a tester of chess engines would buy into this.

I have performed many overnight analysis sessions with many engines, including Rybka and Zappa, and have not noticed any inclination toward 0 scores with increased depth. I doubt my results have any more or less statistical significance than Suj's results.

Alan
Parent - - By turbojuice1122 (Gold) Date 2007-12-17 22:49
The reason that I "buy into this" is because it makes sense based on Zappa's style.  Zappa is not a search workhorse, but instead plays in more of a Deep Blue style (though hundreds of elo stronger).  I would guess that with such a style, there is a range of "critical depths" in which the dominant feature will be something like Rybka in "ultra-pessimistic" mode.  There is also the possibility that something like this would tend to occur more with 10 times the computing power from multiple CPU than with 10 times the game length on "ordinary" quad machines.  This would not occur for all engines, so I'm certainly not using this to establish asymptotic behavior of chess engines in general.  In particular, I would guess that this would be the opposite case for Rybka if she could be scaled well on such machines, since Rybka is a search hog.
Parent - - By NATIONAL12 (Gold) Date 2007-12-17 23:00
why dont you make life much simpler for yourself and buy a cheap quad.
Parent - By turbojuice1122 (Gold) Date 2007-12-18 00:08
I like the computer I have, and I have no room for more stuff.
Parent - - By Banned for Life (Gold) Date 2007-12-17 23:16
I can't follow your argument at all, but the conclusion seems to be that chess is nearly solved (i.e we're close enough to see the draw score at the end of the rainbow) and I would be able to see this too if only I had a 32+ core machine like Suj. I'm not buying it.

Alan
Parent - By turbojuice1122 (Gold) Date 2007-12-18 00:06
No, no, that's not what I'm saying at all: what I'm saying is that Zappa's methods make this seem like the case, but that if Rybka scaled just as well on such systems, she would have a much higher rating because she'd be able to "break through", so to speak, without wasting all kinds of time worrying about strange things that the opponent might come up with that only apply in that critical depth range anyway.  In spite of this, Zappa is still the strongest thing around because it scales very well on 32-core machines and such, and is able to make good enough use of mistakes of "lowly" computers like Rybka on a quad or octo that it demolishes anything around.
Parent - - By suj (***) Date 2007-12-26 02:46 Edited 2007-12-26 02:50
Hi Alan,

Its simple logic when you have 2 huge hardware playing against each of them it results in mostly draws irrespective of the engine u use namely rybka or zappa.I would say zappa is marginally better on bigger hardware in comparison to 232a for simple reason scaling and possibly better endgame which am sure some will debate and time control being 90/30 and above.

When you use big hardware the ply they reach is adequate enough for them to fend of any bad moves or attacks and hence the significant increase in draws(example of that was a koe game last year vs Hiarcs (Kevin Stock) which was run on 64 and zappa was up 1.06 and then drew.That is why after careful experiment not casual zappa as of now is suited/tuned best for max upto 32 cores and anything higher is just waste of energy I would say.In some words it is quite similar to rybka playing better on a highly overclocked quad than an octa.

I did come to this conclusion after running with and without out book games and book is kinda useless if u really use big hardware.

As of now the only use of bigger hardware is surely and definite saves analysis time .... LOL I know a few who have sat on moves for hours if not days for their correspondence games but its just a matter of minutes to hit very deep ply on big hardware on zappa.Actually rybka also attains ply quicker though kns is the big question mark.

Do not compare some Eco as engine dont completely understand chess yet....Dagh I still havent forgotten our B97 game lol(j/k)

Belated Merry Christmas wishes and Happy New Year.

Suj

Hope this puts an end to this debate
Parent - - By Banned for Life (Gold) Date 2007-12-26 05:07
Suj,

The debate must continue! :-)

The Rybka-Zappa match had 50% draws. I suspect if the match had been longer, this percentage would have be higher, but this was the best two engines running on pretty good hardware (octal core machine) for a reasonable period of time. Maybe next year there will be a match on 16 core machines. I don't think this will change the expected draw percentage at all.

In fact, I haven't seen any tendency toward more draws with better engines, longer games, or with more cores, and all statements to the contrary tend to be anecdotal. Playing the same engine against itself generally results in more draws irrespective of engine or hardware, but this is not surprising since an engine is less likely to have a divergent view of a game from its opponent, when its opponent is the same engine.

The higher percent draws with bigger hardware is just another way of saying that the game is almost solved. With engines playing better than the best humans, there is a natural tendency for people to believe that the best engines on the best hardware play nearly perfect chess. This is blatantly false, as will be obvious looking back 10 years from now.

Since I have a very good memory, I will remind you of this conversation 5 years from now when you have your 512 core machine and claim that you are seeing increased draw percentages that the rest of us aren't seeing because we don't have big enough hardware. :-)

Regards,
Alan

None of this should be construed to mean that I wouldn't want to have you run my book on one of your monster machines for the next freestyle! :-)
Parent - - By Uri Blass (*****) Date 2007-12-26 11:40
I think that there are more draws with longer time control.
I decided to do some comparison of spike results in ccrl because this engine played many games.

See the following linkes:

http://computerchess.org.uk/ccrl/404/cgi/engine_details.cgi?print=Details&eng=Spike%201.2%20Turin#Spike_1_2_Turin
http://computerchess.org.uk/ccrl/4040.live/cgi/engine_details.cgi?print=Details&eng=Spike%201.2%20Turin#Spike_1_2_Turin

At 40/4 I get the following results:

9659 games: 2736 wins, 3720 losses, 3203 draws (33.2%), score: 44.9%

At 40/40 I get the following results

4067 games: 1073 wins, 1322 losses, 1672 draws (41.1%), score: 46.9%

It seems that there are more draws at 40/40

The reason could be the choice of the opponents but it seems that it is not the case.
I did some comparison for common opponnents and it seems that against the best opponents and when the number of draws
were the same the percentage of draw was smaller at 40/4 because the number of games at 40/4 was bigger:

40/40
Rybka 2.3.2a 64-bit 2CPU 3099 +21
−21 (+245) 6.5 − 27.5
(+0−21=13) 19.1%

40/4
–   Rybka 2.3.2a 64-bit 2CPU 3102 +13
−13 (+277) 10.5 − 53.5
(+4−47=13) 16.4%

same number of draws more games

40/40

Rybka 2.3.2a 64-bit 3075 +17
−17 (+221) 5 − 38
(+0−33=10) 11.6%

40/4

  Rybka 2.3.2a 64-bit 3073 +14
−14 (+248) 8 − 40
(+3−35=10) 16.7% same number of draws more games

40/40
   Zappa Mexico 64-bit 4CPU 3064 +20
−19 (+210) 16 − 52
(+4−40=24) 23.5%

40/4

  Zappa Mexico 64-bit 4CPU 3040 +13
−13 (+215) 17 − 81
(+5−69=24) 17.3% same number of draws more games.
Parent - By Banned for Life (Gold) Date 2007-12-26 18:02
OK, it looks like I was incorrect and longer time controls are indeed correlated to a higher draw percentage. Next question is whether these percentages are higher than they were 5 years ago for the same time control.

Alan
Parent - By turbojuice1122 (Gold) Date 2007-12-26 17:47
You can also see Uri's point on the CEGT lists, which is the rebuttal I thought of immediately when I saw your quotes: compare especially the 40/20 and 40/120 lists.  Definitely a noticeably higher percentage of engines with over 40% draws on the tournament time control list.

40/20 list

40/120 list.

There is an extreme situation with almost 50% draws on the 40/120 quad list:
40/120 quad
Parent - By BB (****) Date 2007-12-21 00:05
it's one problem is that it searches too hard for opponent refutations, and the result is often a draw.

I agree wholeheartedly that this is likely to be nonsense. If I were to have the data: scales well up to 16 processors, but has little gain from 16-32-64 and almost none from 64-512 (though I'm not sure how large his test sample at 512 really was), I'd strongly suspect some obscure bug (which could be quite difficult to detect and trace, given the perils of multiprocessing).
Parent - - By BB (****) Date 2007-12-17 23:50

> it searches too hard for opponent refutations, and the result is often a draw.


In fact it is now conceded by all experts that by proper play on both sides the legitimate issue of a game ought to be a draw.

> Rybka's style seems to be based on the logic that the enemy is going to make a mistake eventually, all you have to do is wait for this mistake to happen and then just use this mistake to your advantage and win.


We entirely agree with Baron von Heydebrand und der Lasa who lays down the sound maxim: "The simplest and the shortest way of winning is the best."

[Both quotes are from Steinitz on "The Modern School and Its Tendency"]
Parent - By Uly (Gold) Date 2007-12-18 00:04

> We entirely agree with Baron von Heydebrand und der Lasa who lays down the sound maxim: "The simplest and the shortest way of winning is the best."


But Rybka's style won't attain that, a better theoretical approach is to attack the enemy directly, even if you open holes on your position (This is because we assume that the opponent is not good enough and so it won't take advantage of Rybka's weaknesses - While Rybka assumes that she's playing another Rybka and so goes solid instead of aggressive, this will not reach the shortest way to win.)

This aggressive Rybka will not score as good against other Rybkas or Zappa Mexico on 16 cores, so that may be the reason that the solid style will remain.
Parent - - By premraj_n (**) Date 2007-12-18 04:04 Edited 2007-12-18 04:12
Suj said it was draw mostly as it keep on looking for refutations on a 128 cores+ but he accepted its not the case when zappa mexico played on 32cores maximum which were on SAME MOTHERBOARD = 4CPU with 8CORES or even 4CPU with 4CORE which is 16threads where it plays very powerful.

The difference is scaling on same motherboard vs scaling to more clusters.
But my point is when Rybka analyses its depth of thinking just reaches 20+ in just 1min on my quad and when i make a move which is analysed as better, the Rybka didnot give equal or more priority and had only analysed to depth 13. On the other hand if it had analysed like say upto 4variations or even 2 then reaching depth to 20 would mean good analysis by Rybka.
But still because of low kN/s it need lot more time for Rybka compared to other engines that generates more node speed.

For example take the game played in WCCC '07 Rybka vs Shredder and opening was B97 and analyse the moves played by white with Rybka see if Rybka directly chooses
22.Qg3 when Shredder had played 21...f5.
Parent - By turbojuice1122 (Gold) Date 2007-12-18 04:09
That's interesting--so for whatever reason, Zappa plays in something like "ultra-pessimistic" mode when it has more than 32 cores--sounds like something that could potentially be fixed in the programming, but I'm not sure if that wouldn't be harmful to its play on 32 cores or fewer, like what most of us mere mortals have.
Parent - - By Carl Bicknell (*****) Date 2007-12-18 10:23
Well that's something that confuses me. People say that Zappa IS very solid, at least as solid as rybka but you seem to be implying it isn't in your above post where you say its playing style is dfferent to Ryka's.

explanation appreciated!
Parent - - By turbojuice1122 (Gold) Date 2007-12-18 11:02
Zappa is solid, yes, but not as "deep" as Rybka, and more tactical (but not as tactical as Fritz X, where X is less than or equal to 10).
Parent - By Uly (Gold) Date 2007-12-18 11:42
Also, compared to Rybka, it plays more aggressively.
Parent - - By Vasik Rajlich (Silver) Date 2007-12-20 13:39
95% of the work is more accurate evaluation and more efficient search.

We may also include new personalities and a persistent hash.

Vas
Parent - - By maxrom (***) Date 2007-12-21 10:46
I think rybka 2.3 have the best search efficience from all the chess programs, the 2 weakest points of rybka 2.3 were endgame knowlegend and kingssafe . Is this right ?

Why you work on search efficiency when rybka 2.3 have a very good search efficience ?
Parent - - By Vasik Rajlich (Silver) Date 2007-12-25 15:11
Search efficiency and middlegame evaluation are always important, you can never stop working on those.

Vas
Parent - - By maxrom (***) Date 2007-12-25 16:01
Well, endgame knowlegend and kingsattack also very Important and that are the weakest points of rybka 2.3
Parent - - By Uri Blass (*****) Date 2007-12-25 21:21
I do not think that king attack is one of the weakest points of rybka2.3.2a and rybka often win games by king attack

Some examples from CEGT for you

[Event "CEGT Quad tournament time control"]
[Site "CEGT"]
[Date "2007.10.23"]
[Round "6.1"]
[White "Rybka 2.3.2a 64 4CPU"]
[Black "Zappa Mexico X64 4CPU"]
[Result "1-0"]
[ECO "B80"]
[Annotator "0.29;0.27"]
[PlyCount "103"]
[EventDate "2007.??.??"]
[EventRounds "13"]
[TimeControl "40/4800:20/2400:1200"]

{Intel(R) Core(TM)2 Quad CPU @ 2.40GHz 3 MHz W=19.3 ply; 434kN/s; HS-8moves.
ctg B=15.5 ply; 2,306kN/s; 182 TBAs; HS-8moves.ctg} 1. e4 {0} c5 {0} 2. Nf3 {0}
d6 {0} 3. d4 {0} cxd4 {0} 4. Nxd4 {0} Nf6 {0} 5. Nc3 {0} a6 {0} 6. Be3 {0} e6 {
0} 7. f3 {0} b5 {0} 8. Qd2 {0} Nbd7 {0} 9. O-O-O {0.29/17 144} Bb7 {0.27/18 980
} 10. a3 {0.35/18 114} Rc8 {(Ne5) 0.23/17 237} 11. g4 {0.20/18 141} h6 {
(Ne5) 0.19/17 198} 12. Kb1 {(Rg1) 0.21/16 113} Ne5 {0.10/17 784} 13. Qe1 {
(Rg1) 0.25/17 97} d5 {(Be7) -0.11/15 206} 14. Bf4 {(Ndxb5) 0.54/17 108} Ng6 {
0.18/16 149} 15. Bg3 {0.46/18 81} Qb6 {(Bc5) 0.10/16 178} 16. Bd3 {
(Ndxb5) 0.36/17 90} Bc5 {0.19/16 90} 17. Nb3 {0.38/18 75} Be7 {0.28/16 85} 18.
e5 {0.46/19 91} Nd7 {0.31/18 250} 19. f4 {(Bxg6) 0.46/19 86} d4 {
(0-0) 0.53/16 105} 20. Ne4 {(Ne2) 0.61/18 125} Bd5 {(b4) 0.64/15 104} 21. h4 {
(Qf2) 0.85/18 135} b4 {0.64/14 75} 22. a4 {0.90/17 33} Nc5 {0.74/15 87} 23.
Nbxc5 {0.85/17 86} Bxc5 {0.74/14 0} 24. f5 {0.88/18 41} b3 {0.65/15 93} 25. Rh2
{0.70/18 316} Nf8 {(Ne7) 0.84/15 143} 26. Nd6+ {0.76/18 122} Bxd6 {0.84/14 0}
27. exd6 {0.74/19 37} h5 {(Nd7) 0.84/15 86} 28. g5 {(a5) 1.13/16 220} Nd7 {
0.82/16 80} 29. g6 {(a5) 1.01/17 47} e5 {1.05/15 106} 30. Bxe5 {1.19/18 73} O-O
{1.06/16 65} 31. f6 {1.35/17 76} fxg6 {1.27/15 179} 32. Bxg6 {(Qg3) 1.36/18 121
} Nxf6 {0.93/15 57} 33. Rxd4 {1.28/18 85} Bc4 {(Qc6) 1.01/14 52} 34. Rg2 {
(Bd3) 1.24/17 60} Ng4 {0.97/15 90} 35. Rg1 {1.97/17 157} Bf7 {
(bxc2+) 1.12/14 61} 36. Bd3 {2.02/15 0} Rfd8 {(bxc2+) 2.06/14 87} 37. c3 {
3.15/21 430} Qc5 {2.70/14 34} 38. Qe4 {(Rgxg4) 4.33/22 239} Qxe5 {4.10/14 22}
39. Qh7+ {3.98/20 0} Kf8 {4.10/13 0} 40. Qh8+ {4.58/21 89} Bg8 {4.10/13 0} 41.
Rdxg4 {4.58/22 74} hxg4 {4.10/13 0} 42. Rf1+ {4.73/22 45} Ke8 {4.10/13 0} 43.
Bf5 {4.90/22 74} Rxd6 {4.10/13 0} 44. Qxg8+ {4.97/21 30} Ke7 {4.10/13 0} 45.
Qxc8 {5.31/22 252} Rf6 {4.10/13 0} 46. Ka1 {5.31/20 72} g3 {6.48/16 264} 47.
Rd1 {(Qb7+) 5.27/19 30} Rxf5 {2.94/11 1} 48. Rd7+ {5.41/19 45} Kf6 {2.94/11 0}
49. Qxa6+ {5.55/18 25} Qe6 {2.94/11 0} 50. Rd6 {5.61/18 49} Qxd6 {
(Re5) 6.76/19 401} 51. Qxd6+ {4.63/16 0} Kf7 {6.76/18 0} 52. Qd1 {5.95/17 156}
1-0

[Event "CEGT Quad tournament time control"]
[Site "CEGT"]
[Date "2007.10.25"]
[Round "3.1"]
[White "Rybka 2.3.2a 64 4CPU"]
[Black "Naum 2.2 x64 4CPU"]
[Result "1-0"]
[ECO "C18"]
[Annotator "0.00;0.84"]
[PlyCount "65"]
[EventDate "2007.??.??"]
[EventRounds "13"]
[TimeControl "40/4800:20/2400:1200"]

{Intel(R) Core(TM)2 Quad CPU @ 2.40GHz 2400 MHz W=20.4 ply; 387kN/s; HS-8moves.
ctg B=21.6 ply; 4.183kN/s; HS-8moves.ctg} 1. e4 {-300.02/0 0} e6 {-300.02/0 0}
2. d4 {-300.02/0 0} d5 {-300.02/0 0} 3. Nc3 {-300.02/0 0} Bb4 {-300.02/0 0} 4.
e5 {-300.02/0 0} c5 {-300.02/0 0} 5. a3 {-300.02/0 0} Bxc3+ {-300.02/0 0} 6.
bxc3 {-300.02/0 0} Ne7 {-300.02/0 0} 7. Qg4 {-300.02/0 0} Qc7 {-300.02/0 0} 8.
Qxg7 {-300.02/0 0} Rg8 {-300.02/0 0} 9. Qxh7 {0.00/20 126} cxd4 {0.84/21 375}
10. Ne2 {0.07/20 122} dxc3 {(Sbc6) 0.71/20 143} 11. f4 {(h4) 0.23/19 64} Bd7 {
0.71/19 143} 12. Nd4 {(Tb1) 0.00/20 995} a6 {0.02/20 142} 13. Be3 {0.21/20 75}
Nbc6 {0.00/21 143} 14. Rb1 {(g3) 0.09/19 95} Nxd4 {(Da5) 0.00/19 142} 15. Bxd4
{0.32/19 43} Bb5 {(Da5) 0.00/21 142} 16. Kf2 {(g3) 0.21/17 28} Kf8 {
(Lxf1) 0.65/19 380} 17. Bxb5 {0.49/18 27} Nf5 {0.85/20 132} 18. Rhd1 {
0.60/19 34} axb5 {1.30/20 265} 19. Rxb5 {0.68/20 36} Nxd4 {1.24/20 127} 20.
Rxd4 {0.98/20 72} Rxa3 {1.24/20 126} 21. Rdb4 {1.07/20 36} Ra1 {1.22/20 127}
22. Qh4 {(g3) 1.23/20 164} b6 {(Ke8) 1.65/19 253} 23. f5 {1.43/19 54} Ra6 {
1.46/18 119} 24. Qf6 {1.38/19 33} Qe7 {1.45/20 120} 25. Qh6+ {(h4) 2.10/21 37}
Ke8 {1.74/21 239} 26. f6 {1.90/20 0} Qd8 {2.00/21 279} 27. Qe3 {1.90/18 0} Qc7
{(Th8) 2.57/21 199} 28. Rxb6 {3.79/22 41} Rxb6 {2.68/21 92} 29. Rxb6 {3.55/21 0
} Rg4 {(d4) 2.67/21 92} 30. Ra6 {5.29/21 45} Re4 {(Dc4) 2.88/21 230} 31. Ra8+ {
4.63/19 0} Kd7 {11.10/26 339} 32. Ra7 {4.63/18 0} Rxe3 {14.37/25 197} 33. Rxc7+
{7.14/17 34} 1-0

[Event "CEGT Quad tournament time control"]
[Site "CEGT"]
[Date "2007.10.25"]
[Round "4.1"]
[White "Rybka 2.3.2a 64 4CPU"]
[Black "Naum 2.2 x64 4CPU"]
[Result "1-0"]
[ECO "C18"]
[Annotator "0.18;0.15"]
[PlyCount "95"]
[EventDate "2007.??.??"]
[EventRounds "13"]
[TimeControl "40/4800:20/2400:1200"]

{Intel(R) Core(TM)2 Quad CPU @ 2.40GHz 2610 MHz W=19.8 ply; 410kN/s; HS-8moves.
ctg B=21.2 ply; 4,794kN/s; HS-8moves.ctg} 1. e4 {-300.02/0 0} e6 {-300.02/0 0}
2. d4 {-300.02/0 0} d5 {-300.02/0 0} 3. Nc3 {-300.02/0 0} Bb4 {-300.02/0 0} 4.
e5 {-300.02/0 0} c5 {-300.02/0 0} 5. a3 {-300.02/0 0} Ba5 {-300.02/0 0} 6. b4 {
-300.02/0 0} cxd4 {-300.02/0 0} 7. Nb5 {-300.02/0 0} Bc7 {-300.02/0 0} 8. f4 {
-300.02/0 0} Bd7 {-300.02/0 0} 9. Nxc7+ {0.18/18 130} Qxc7 {0.15/21 150} 10.
Nf3 {0.23/19 181} Nh6 {(La4) 0.28/21 150} 11. Bd3 {0.22/20 114} Nc6 {
0.46/20 150} 12. O-O {(Lb2) 0.23/19 574} O-O {(Sg4) 0.56/19 150} 13. Ng5 {
(Lb2) 0.19/18 111} g6 {0.51/19 150} 14. Qe1 {(g4) 0.25/19 125} Ng4 {
(Sf5) 0.50/19 150} 15. Nxh7 {0.43/18 47} Kxh7 {0.61/21 166} 16. Qh4+ {
0.46/19 77} Kg7 {0.71/22 155} 17. Qxg4 {0.47/20 33} a6 {(Th8) 0.89/21 149} 18.
Bd2 {(Lb2) 0.59/20 461} Rh8 {(f5) 0.57/19 149} 19. Be1 {0.58/18 92} Qd8 {
0.93/20 310} 20. Bf2 {0.63/18 90} Rc8 {(b5) 0.93/19 141} 21. Rfc1 {
(Tfd1) 0.63/16 116} Rh6 {(Th5) 0.92/20 141} 22. h3 {(Dg3) 0.65/17 220} f6 {
(Th5) 0.91/19 141} 23. Qg3 {(Tf1) 0.83/20 226} Kh8 {(De7) 0.71/20 141} 24. c3 {
0.88/18 71} fxe5 {(dxc3) 0.80/19 187} 25. fxe5 {1.11/19 85} dxc3 {1.31/22 277}
26. Rxc3 {1.20/19 27} Qe7 {1.34/21 235} 27. Rf1 {(Le3) 1.15/19 54} Rg8 {
1.93/21 243} 28. Rcc1 {(Le2) 1.20/18 154} Be8 {1.17/19 112} 29. Be2 {
(Lc5) 1.18/18 69} g5 {1.30/20 112} 30. Bc5 {1.27/20 137} Qg7 {1.43/20 112} 31.
Bd6 {(Df2) 1.27/19 41} Bg6 {1.45/19 241} 32. Rf6 {1.61/19 227} Qh7 {1.56/19 115
} 33. Rf8 {(Lf8) 1.69/18 41} Be4 {2.07/20 192} 34. Rxg8+ {(Df2) 1.73/19 61}
Qxg8 {1.94/20 82} 35. Bxa6 {(Df2) 1.62/19 0} Nd4 {1.75/20 138} 36. Qf2 {
1.81/21 80} bxa6 {2.12/20 142} 37. Qxd4 {1.92/22 82} Rh7 {2.26/20 54} 38. Qc5 {
1.95/22 238} Qe8 {2.26/20 53} 39. Qb6 {2.19/21 48} Bd3 {2.36/20 54} 40. Qe3 {
1.98/19 0} Bc4 {2.36/22 54} 41. Qxg5 {1.98/17 0} Rh5 {2.49/22 120} 42. Qe3 {
(Df6+) 2.28/20 36} Kh7 {(Df7) 2.90/20 240} 43. Qa7+ {3.13/22 107} Kh6 {
(Kg8) 3.84/22 339} 44. Rc3 {3.99/18 23} Rf5 {4.09/21 249} 45. Rg3 {3.61/16 0}
Qf7 {4.31/21 226} 46. Be7 {4.92/19 58} Bb5 {(d4) 5.93/21 326} 47. Qe3+ {
4.45/17 0} Kh7 {7.12/22 192} 48. Bf6 {5.65/17 61} 1-0

[Event "CEGT Quad tournament time control"]
[Site "CEGT"]
[Date "2007.10.26"]
[Round "6.1"]
[White "Rybka 2.3.2a 64 4CPU"]
[Black "Naum 2.2 x64 4CPU"]
[Result "1-0"]
[ECO "C11"]
[Annotator "0.11;-0.16"]
[PlyCount "77"]
[EventDate "2007.??.??"]
[EventRounds "13"]
[TimeControl "40/4800:20/2400:1200"]

{Intel(R) Core(TM)2 Quad CPU @ 2.40GHz 2610 MHz W=20.6 ply; 431kN/s; HS-8moves.
ctg B=22.0 ply; 4,417kN/s; HS-8moves.ctg} 1. e4 {-300.02/0 0} e6 {-300.02/0 0}
2. d4 {-300.02/0 0} d5 {-300.02/0 0} 3. Nc3 {-300.02/0 0} Nf6 {-300.02/0 0} 4.
e5 {-300.02/0 0} Nfd7 {-300.02/0 0} 5. f4 {-300.02/0 0} c5 {-300.02/0 0} 6. Nf3
{-300.02/0 0} Nc6 {-300.02/0 0} 7. Be3 {-300.02/0 0} a6 {-300.02/0 0} 8. Qd2 {
-300.02/0 0} b5 {-300.02/0 0} 9. a3 {0.11/16 124} Qb6 {(Lb7) -0.16/19 150} 10.
Be2 {0.14/17 102} Be7 {-0.20/19 150} 11. dxc5 {(0-0) 0.11/19 204} Bxc5 {
-0.15/21 150} 12. Bxc5 {0.10/20 67} Qxc5 {(Sxc5) -0.05/21 230} 13. b4 {
(Td1) 0.11/19 184} Qe7 {(Db6) 0.20/21 162} 14. O-O {0.30/19 68} Bb7 {
0.40/20 436} 15. Rab1 {(a4) 0.33/18 29} O-O {0.37/19 135} 16. a4 {0.30/19 53}
Nxb4 {(bxa4) 0.52/20 135} 17. axb5 {0.35/17 41} a5 {(Dc5+) 0.62/21 135} 18. Nd4
{0.41/16 65} Nc5 {0.67/20 135} 19. Qe3 {0.66/17 33} Rfc8 {0.47/20 135} 20. f5 {
(Tb2) 0.45/18 221} Ne4 {0.66/19 135} 21. Nxe4 {0.48/18 42} dxe4 {0.60/20 135}
22. f6 {(c4) 0.84/20 241} Qc5 {0.13/19 135} 23. Qg5 {0.91/20 70} g6 {
1.01/21 338} 24. Kh1 {1.07/19 37} Qf8 {0.38/21 308} 25. Nxe6 {1.13/20 62} fxe6
{1.08/23 280} 26. f7+ {0.85/21 94} Kh8 {1.12/22 101} 27. Rbd1 {1.59/21 203} Nd5
{1.80/21 264} 28. c4 {1.44/19 0} Qe7 {1.80/22 149} 29. Qxe7 {1.79/23 103} Nxe7
{1.80/20 1} 30. Rd7 {2.10/24 64} Nf5 {2.52/23 226} 31. Rxb7 {2.14/24 79} Nd4 {
2.73/22 77} 32. Bg4 {(Ld1) 2.23/23 70} h5 {(Tf8) 2.03/21 77} 33. Rd7 {
3.33/24 103} hxg4 {3.33/23 231} 34. b6 {3.07/22 0} Rab8 {3.34/23 55} 35. Rxd4 {
3.07/20 0} Rf8 {3.56/23 55} 36. c5 {3.11/19 0} e3 {3.77/22 55} 37. Re4 {
(Kg1) 3.96/22 545} Kg7 {5.03/23 110} 38. Rxe3 {3.58/19 0} a4 {(Tfc8) 6.87/22 55
} 39. Kg1 {(Ta3) 5.88/16 67} 1-0

[Event "CEGT Quad tournament time control"]
[Site "CEGT"]
[Date "2007.10.26"]
[Round "1.1"]
[White "Rybka 2.3.2a 64 4CPU"]
[Black "Naum 2.2 x64 4CPU"]
[Result "0-1"]
[ECO "C13"]
[Annotator "0.20;0.08"]
[PlyCount "68"]
[EventDate "2007.??.??"]
[EventRounds "13"]
[TimeControl "40/4800:20/2400:1200"]

{Intel(R) Core(TM)2 Quad CPU @ 2.40GHz 2400 MHz W=20.3 ply; 464kN/s; HS-8moves.
ctg B=21.0 ply; 4.620kN/s; HS-8moves.ctg} 1. d4 {0} e6 {0} 2. e4 {0} d5 {0} 3.
Nc3 {0} dxe4 {0} 4. Nxe4 {0} Nd7 {0} 5. Nf3 {0} Ngf6 {0} 6. Bg5 {0} Be7 {0} 7.
Nxf6+ {0} Bxf6 {0} 8. Bxf6 {0} Nxf6 {0} 9. Bc4 {0.20/19 110} O-O {0.08/22 150}
10. O-O {0.16/19 62} b6 {0.09/21 150} 11. Ne5 {0.14/20 213} Bb7 {0.01/20 150}
12. Re1 {0.19/20 965} Qe7 {(Dd6) 0.12/19 150} 13. Nxf7 {(Dd3) 0.27/19 157} Kxf7
{(Txf7) -0.59/21 150} 14. Bxe6+ {(Txe6) 0.28/18 22} Qxe6 {-0.50/23 150} 15.
Rxe6 {0.27/20 74} Kxe6 {-0.50/21 150} 16. Qd3 {0.29/20 54} Kf7 {-0.55/18 150}
17. Qc4+ {(Te1) 0.33/21 197} Nd5 {-0.58/20 150} 18. Re1 {0.33/21 67} Rad8 {
-0.52/19 168} 19. g3 {(Da4) 0.32/21 636} Kg8 {(Tfe8) -0.56/19 149} 20. f4 {
0.36/17 49} Rd7 {(Tf7) -0.75/20 149} 21. Qb3 {(c3) 0.45/17 102} Rfd8 {
(h6) -0.83/20 149} 22. c3 {0.33/18 159} h5 {(a5) -0.99/21 149} 23. h4 {
(Da4) 0.42/18 148} Kh8 {(Tf7) -1.40/22 204} 24. Qa4 {0.21/20 205} Nf6 {
-1.58/22 146} 25. Qxa7 {0.00/20 36} Ba8 {(Le4) -1.85/22 146} 26. Kf1 {
(Da4) -0.25/20 338} Be4 {-2.84/20 146} 27. Qa6 {-0.29/20 24} c5 {
(Ta8) -3.14/21 146} 28. Ke2 {(Dxb6) -0.69/19 66} cxd4 {-3.38/21 146} 29. c4 {
-1.05/20 52} b5 {(Td6) -3.65/21 146} 30. Kd2 {-2.25/20 232} Rc7 {-4.10/20 146}
31. Qa5 {(c5) -2.98/19 180} Rdc8 {-5.12/21 146} 32. Rxe4 {(Dxb5) -4.46/20 140}
Nxe4+ {-4.65/6 0} 33. Ke2 {(Kd3) -5.21/20 112} Rc5 {-7.63/20 164} 34. Kf3 {
(cxb5) -7.03/18 87} bxc4 {-11.97/21 164} 0-1
Parent - By maxrom (***) Date 2008-01-05 17:01
And rybka lose many game because of kingssafty

2 Exampels are posteded in weakest points of rybka 2.3 by the user strongholder on the first page

You can see here :http://rybkaforum.net/cgi-bin/rybkaforum/topic_show.pl?tid=2629;pg=1
Parent - - By Roland Rösler (****) Date 2007-12-26 02:41
Vas tells us always the same. Search efficiency and middlegame evaluation are important. Here you make the Elo-points. Endgame knowlegend and kingsattack are also very Important . But here you don´t make so much Elo-points. Rybka has to be the best engine in the rating lists. So what is to do? It´s your task to convince Vas. I can´t, because I´m not convinced about the right way for this rating list issue.
Parent - - By M ANSARI (*****) Date 2007-12-26 07:12
Personally, I think that by far the most important improvement in Rybka 3 would be MP efficiency.  Rybka loses too much in multi core scaling when compared to the likes of Zappa.  With stronger hardware ... things like king attacks and endgame play should automatically improve.  8 cores would seem like the average hardware in chess engine tournaments in the coming year, therefore having an engine that can efficiently use those 8 cores will be vital to keep Rybka's big edge.
Parent - - By Bouddha (****) Date 2007-12-26 08:54
Hi !

While I can agree that in future most of us will use 8 cores or more, I currently for 2008 havent hird 8 cores comming and thus we will need to have motherboard which can have 2 cpu which is very few people.

Lets say that 8cpu die comes in around 12 months, that would mean for Rybka 4. Why should we already have something that will benefit only in 1 Year when anyway, there will be a new version (4) of Rybka ?

Best regards
Parent - - By M ANSARI (*****) Date 2007-12-26 13:31
Yes ofcourse the 8 core is not for everyone ... what I mean is that in engine tournaments such as Lieden ... the 8 core will be average hardware.  For Rybka to do well it needs to play well on 8 cores in these tournaments.  For us a Rybka even on 200 Mhz will still score very high.
Parent - - By Bouddha (****) Date 2007-12-26 16:24
Understood and agreed that for computer tournament they may need good 8cpu or more and an engine that scale good on that.

Nevertheless Rybka is a commercial engine and the current field uses mainly 2 or 4 cpu.

Therefore I personnally think the priority should be put on :
- endgame knowledge
- tacticks abality

regards
Parent - By maxrom (***) Date 2008-01-05 15:08
And kingssatack and safty
Parent - By staylor (***) Date 2008-01-06 14:59
Why would he work on search efficiency when that is already good?

Well, Vas himself already answered when he explained you never stop working on that. He answered the correct answer cooly.
I myself nearly exploded! Unless you are trying to say that we should now stop improving chess playing computer strength, and just work on making end-game/opening better. And that's the end of the story.
Parent - - By Vempele (Silver) Date 2008-01-06 15:20

> Why you work on search efficiency when rybka 2.3 have a very good search efficience ?


It's only 'good' in a relative sense. Not really even comparable to humans.
Parent - - By maxrom (***) Date 2008-01-06 15:35
I'm right when I say that seearch efficiency isn't 1 of the weakest points in rybka 2.3 ???

I'm right when I say that the weakest poins in 2.3 is kingssatack ( safety )  and endgame knowlegend ???
Parent - By Uri Blass (*****) Date 2008-01-06 15:58
An extreme example:
Making rybka 100 times faster that is only search efficiency is going to give rybka more rating then king safety or endgame knowledge.

Maybe Vas cannot do it but you do not know what he can do so you have no basis for your conclusions.
Parent - - By vroger007 (**) Date 2008-01-04 18:16
Hi Vas,

What about EGTB handling and "BU" in 3.0 ?

Roger
Parent - - By Vasik Rajlich (Silver) Date 2008-01-05 13:57
Roger,

ok, you got me - what is BU?

Vas
Parent - - By Debaser (***) Date 2008-01-05 15:00
My bet, bishop underpromotion.

You promised it for the last Rybka 2.x!!! ;)
Parent - - By vroger007 (**) Date 2008-01-05 18:29
correct :-)
Up Topic Rybka Support & Discussion / Rybka Discussion / New improvements in Rybka 3
1 2 Previous Next  

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill