GM Dzindzichashvili - Rybka, 4-4

[Henrik Dinesen]

Obviously a Chess variant.....

Yes, but within reason...While reason is there, it's never gonna be the kind of games we collect, and for me, it's not some I can put into my DB for reference...
Anyway, I respect the goal, to level things a bit... and it seem to work to extent.

[Mike S.]

If the line 1.e4 e5 2.f4 exf4 is being played because both 2 players have chosen to play it and not because they are forced by the rules of the match to play it, then it is Chess. (...) In these games Rybka played with Roman, this freedom wasn't there! Obviously a Chess variant.....

This is your very individual opinion.

What about theme tournaments with predefined openings? By your definition, that's not chess either. Example: L'Ami - Stellwagen, March 3rd-6th 2008. They played 4 games of advanced (or 'complete') chess, but predefined was the 2 knights defense, each. Was that NOT chess?!

http://www.chessevents.nl/ast_match.shtml
http://www.chessevents.nl/ast_reports.shtml

I must discard the idea that it wasn't chess but a variant or whatever... forget it. They have played chess.

[Dr.Wael Deeb]

If the line 1.e4 e5 2.f4 exf4 is being played because both 2 players have chosen to play it and not because they are forced by the rules of the match to play it, then it is Chess. (...) In these games Rybka played with Roman, this freedom wasn't there! Obviously a Chess variant.....

This is your very individual opinion.

What about theme tournaments with predefined openings? By your definition, that's not chess either. Example: L'Ami - Stellwagen, March 3rd-6th 2008. They played 4 games of advanced (or 'complete') chess, but predefined was the 2 knights defense, each. Was that NOT chess?!

http://www.chessevents.nl/ast_match.shtml
http://www.chessevents.nl/ast_reports.shtml

I must discard the idea that it wasn't chess but a variant or whatever... forget it. They have played chess.

This is a classical game of chess playing the two knights defence from a classical start position....no one on earth can claim the opposite

BTW,the odds tournaments reminds me of a race between two horses with one of them racing with three legs only because he is much faster than the other one running with four legs....the anology is a little bit rough but you got my point

[playjunior]

I think we a fast machine and a good engine you will be surprised with the results. Yes engines would make mistakes. But not more than what top humans make in endgames.... Perhaps less.

I strongly disagree. In a technical endgame when there is no time trouble, an engine is not comparable to a top GM. And-hardware makes huge difference when it comes to rating on playchess, but for analyzing endgames +3-4 plies deep makes almost no difference.

Recent example was Rybka evaluating an endgame with a white-colored white bishop, white and black pawns on h-file and black king on h8 as +-4.50-or-something like that.

Such things happen. But you should see not only the tree but the forest.
I mean you should look at many many positions of how engines do in endgames and how humans do.
I think for about the last 1.5 year we are at the point that search depths of engines make them comparably strong with top humans at endgames.....
The key here is fast and latest hardware. QUAD computers.

I disagree.
Computers completely miss quiet moves in endgames. Without those, they will never be able to convert a moderate endgame advantage into a win.
Computers like to play forced variations. Technical endgames are about playing a forced variation in the situation when it is most effective. In Russian this is called "ne speshit#" (do not rush or hurry).

Tablebases might help in this particular case, but you know better than me that they are not a solution-you get a huge speed decrease because of them and it doesn't make real difference in average.

I don't think there is a huge decrease in speed.
I also think that there is a noticeable difference from using the endgame tablebases, for some engines anyway.

Tablebase access slows down engine search significantly, correct me someone if I am wrong. If I am not mistaken, it has been shown that the rating of an engine remains around the same when you install 5-men, (although-3 and 4-men are helpful).

Kramnik-Fritz games from Bahrain match are a thrilling piece of example of misplayed endgames by an engine. I don't think that there has been a remarkable advance in that aspect of play since then.

5.5 years from then. In these 5.5 years the advance on hardware was very big but the advance of software was also, especially in last 2 years, very big.

[/quote]

I suggest we have a look to Fritz-Kramnik endgames from 5.5 years back.

This is the final position of Bahrain match game 1. My computer is Core 2 Duo 2.13 Ghz. Fritz operator proposed a draw (because its a dead-draw ) despite Fritz being happy.

Comp Deep Fritz - Kramnik,V, Bahrain Brains m
[d]4k3/2p5/1p2p1p1/p1p1P1P1/P1P3p1/1P6/5PK1/8 b - - 0 1

Code: Select all

Analysis by Toga II 1.4 beta5c:
...
  +/=  (0.62)   Depth: 20/32   00:00:13  40940kN
28...Kd7 29.Kg3 Ke8 30.Kxg4 Kd8 31.Kf3 Kd7 32.Ke4 Kc6 33.Kf4 Kd7 34.Kg3 Ke8 35.Kg2 Kf7 36.Kf3 Ke8 37.Ke3 Ke7 38.f4 Kd7 
  +/=  (0.62)   Depth: 21/32   00:00:23  70620kN
28...Kd7 29.Kg3 Ke8 30.Kxg4 Kd8 31.Kf3 Kd7 32.Ke4 Kc6 33.Ke3 Kd7 34.Ke2 Ke8 35.Kd3 Ke7 36.Ke4 Kd7 37.Kf3 Ke7 38.Kg3 Kd7 39.Kf4 Ke7 
  +/=  (0.62)   Depth: 22/35   00:00:42  127mN
28...Kd7 29.Kg3 Ke8 30.Kxg4 Kd8 31.Kf3 Kd7 32.Ke4 Kc6 33.Kf4 Kd7 34.Kg3 Kc6 35.Kg2 Kb7 36.Kf3 Kc6 37.Ke4 Kd7 38.Ke3 Ke7 39.f4 Kd7 
  +/=  (0.62)   Depth: 23/35   00:01:15  224mN

(,  08.03.2008)

And so on.

Comp Deep Fritz - Kramnik,V, Bahrain Brains m
[d]4k3/2p5/1p2p1p1/p1p1P1P1/P1P3p1/1P6/5PK1/8 b - - 0 1

Code: Select all

Analysis by Fritz 10:
...
  +/-  (0.71)   Depth: 17/31   00:00:01  1180kN
28...c6 29.Kg3 Kd7 30.Kxg4 Kc8 31.Kf4 Kb8 32.Kf3 Ka8 33.Ke3 Kb7 
  +/-  (0.71)   Depth: 18/40   00:00:01  2309kN
28...c6 29.Kg3 Kd7 30.Kxg4 Kc8 31.Kf4 Kb8 32.Kf3 Kc8 33.Ke2 Kc7 34.Kd2 
  +/-  (0.71)   Depth: 19/43   00:00:04  5233kN
28...c6 29.Kg3 Kd7 30.Kxg4 Kc8 31.Kf4 Kb8 32.Kf3 Kc8 33.Ke2 Kd7 34.f4 Ke7 35.Ke3 Kd7 36.Ke4 
  +/-  (0.71)   Depth: 20/44   00:00:07  9413kN
28...c6 29.Kg3 Kd7 30.Kxg4 Kc8 31.Kf4 Kd7 32.Ke3 Kc7 33.Kd3 Kb7 34.f4 Kc7 35.Ke3 Kd7 36.Ke2 
  +/-  (0.71)   Depth: 21/45   00:00:16  22728kN

(,  08.03.2008)

Fruit is also reporting +=0.62.
I doubt that a QUAD would help btw.


Game 2:

[Event "Bahrain Brains m"]
[Site "Manama"]
[Date "2002.10.04"]
[Round "2"]
[White "Kramnik, Vladimir"]
[Black "Comp Deep Fritz"]
[Result "1-0"]
[ECO "D27"]
[WhiteElo "2807"]
[PlyCount "113"]

1. d4 d5 2. c4 dxc4 3. Nf3 Nf6 4. e3 e6 5. Bxc4 c5 6. O-O a6 7. dxc5 Qxd1 8.
Rxd1 Bxc5 9. Kf1 b5 10. Be2 Bb7 11. Nbd2 Nbd7 12. Nb3 Bf8 13. a4 b4 14. Nfd2
Bd5 15. f3 Bd6 16. g3 e5 17. e4 Be6 18. Nc4 Bc7 19. Be3 a5 20. Nc5 Nxc5 21.
Bxc5 Nd7 22. Nd6+ Kf8 23. Bf2 Bxd6 24. Rxd6 Ke7 25. Rad1 Rhc8 26. Bb5 Nc5 27.
Bc6 Bc4+ 28. Ke1 Nd3+ 29. R1xd3 Bxd3 30. Bc5 Bc4 31. Rd4+ Kf6 32. Rxc4 Rxc6 33.
Be7+ Kxe7 34. Rxc6 Kd7 35. Rc5 f6 36. Kd2 Kd6 37. Rd5+ Kc6 38. Kd3 g6 39. Kc4
g5 40. h3 h6 41. h4 gxh4 42. gxh4 Ra7 43. h5 Ra8 44. Rc5+ Kb6 45. Rb5+ Kc6 46.
Rd5 Kc7 47. Kb5 b3 48. Rd3 Ra7 49. Rxb3 Rb7+ 50. Kc4 Ra7 51. Rb5 Ra8 52. Kd5
Ra6 53. Rc5+ Kd7 54. b3 Rd6+ 55. Kc4 Rd4+ 56. Kc3 Rd1 57. Rd5+ 1-0



[d]r7/4kppp/2R5/p3p3/Pp2P3/5PP1/1P5P/4K3 b - - 0 34
Fritz 10 says this is +=0.35 at depth 18. Kramnik said about this position that it is nearly lost for black

Now, a couple of fine examples of waiting-and torturing strategy by Kramnik:

[d]r7/6pp/3k1p2/p1R1p3/Pp2P3/5PP1/1P1K3P/8 w - - 0 37

Rd5+! Computer prefers here Rb5. Kramnik plays Rd5+: he can go Rb5 whenever he wants, but there is difference whether you have your King on d6 or in c6(e6).

Now we are here:
[d]r7/7p/2k2p2/p2Rp1p1/PpK1P3/5PP1/1P5P/8 w - - 0 40
My Fritz 10 already realizes that black is bad, but as white it would never see the move 40. h3! The point is black cannot do anything, and h3 improves, waits, doesn't miss or spoil anything...its superior to rushing out with Kd3 as my Fritz suggests

This one is my favorite in this game.
[d]8/r7/2k2p1p/p2Rp3/PpK1P2P/5P2/1P6/8 w - - 0 43
Kramnik plays 43 h5!- a move that all commentators said is clearly superior to Rd8 that comps are suggesting with huge evals. My Fritz is quite happy with Rd8 with a +1.5 score.

Kramnik,V - Comp Deep Fritz, Bahrain Brains m
[d]8/r7/2k2p1p/p2Rp3/PpK1P2P/5P2/1P6/8 w - - 0 1

Code: Select all

Analysis by Fritz 10:
...
  +-  (1.54)   Depth: 19/46   00:00:38  68024kN
43.Rd8 Rc7 44.h5 f5 45.Rd5 Kb6+ 46.Kb3 fxe4 47.fxe4 Re7 48.Rb5+ Ka6 49.Kc4 Re8 50.Rd5 Re6 51.Rc5 
  +-  (1.48)   Depth: 20/65   00:01:19  139mN

(,  08.03.2008)

Fruit suggests a pointless move-Kd3 here.
Toga likes Rd8, but with a lesser score than Fritz.

This game is ruthless. If you replay it, you can see that after some point Fritz did not have a single counter-chance, a point where it could breathe. Play any alternative line suggested by computer-you will never get a position as cristal-winning as what Kramnik got.

[Henrik Dinesen]

If the line 1.e4 e5 2.f4 exf4 is being played because both 2 players have chosen to play it and not because they are forced by the rules of the match to play it, then it is Chess. (...) In these games Rybka played with Roman, this freedom wasn't there! Obviously a Chess variant.....

This is your very individual opinion.

What about theme tournaments with predefined openings? By your definition, that's not chess either. Example: L'Ami - Stellwagen, March 3rd-6th 2008. They played 4 games of advanced (or 'complete') chess, but predefined was the 2 knights defense, each. Was that NOT chess?!

http://www.chessevents.nl/ast_match.shtml
http://www.chessevents.nl/ast_reports.shtml

I must discard the idea that it wasn't chess but a variant or whatever... forget it. They have played chess.

This is a classical game of chess playing the two knights defence from a classical start position....no one on earth can claim the opposite

BTW,the odds tournaments reminds me of a race between two horses with one of them racing with three legs only because he is much faster than the other one running with four legs....the anology is a little bit rough but you got my point

The royal garden is playing something symphonic, and the boys band plays techno?

[Uri Blass]

I think we a fast machine and a good engine you will be surprised with the results. Yes engines would make mistakes. But not more than what top humans make in endgames.... Perhaps less.

I strongly disagree. In a technical endgame when there is no time trouble, an engine is not comparable to a top GM. And-hardware makes huge difference when it comes to rating on playchess, but for analyzing endgames +3-4 plies deep makes almost no difference.

Recent example was Rybka evaluating an endgame with a white-colored white bishop, white and black pawns on h-file and black king on h8 as +-4.50-or-something like that.

Such things happen. But you should see not only the tree but the forest.
I mean you should look at many many positions of how engines do in endgames and how humans do.
I think for about the last 1.5 year we are at the point that search depths of engines make them comparably strong with top humans at endgames.....
The key here is fast and latest hardware. QUAD computers.

I disagree.
Computers completely miss quiet moves in endgames. Without those, they will never be able to convert a moderate endgame advantage into a win.
Computers like to play forced variations. Technical endgames are about playing a forced variation in the situation when it is most effective. In Russian this is called "ne speshit#" (do not rush or hurry).

Tablebases might help in this particular case, but you know better than me that they are not a solution-you get a huge speed decrease because of them and it doesn't make real difference in average.

I don't think there is a huge decrease in speed.
I also think that there is a noticeable difference from using the endgame tablebases, for some engines anyway.

Tablebase access slows down engine search significantly, correct me someone if I am wrong. If I am not mistaken, it has been shown that the rating of an engine remains around the same when you install 5-men, (although-3 and 4-men are helpful).

Kramnik-Fritz games from Bahrain match are a thrilling piece of example of misplayed endgames by an engine. I don't think that there has been a remarkable advance in that aspect of play since then.

5.5 years from then. In these 5.5 years the advance on hardware was very big but the advance of software was also, especially in last 2 years, very big.

I suggest we have a look to Fritz-Kramnik endgames from 5.5 years back.

This is the final position of Bahrain match game 1. My computer is Core 2 Duo 2.13 Ghz. Fritz operator proposed a draw (because its a dead-draw ) despite Fritz being happy.

Comp Deep Fritz - Kramnik,V, Bahrain Brains m
[d]4k3/2p5/1p2p1p1/p1p1P1P1/P1P3p1/1P6/5PK1/8 b - - 0 1

Code: Select all

Analysis by Toga II 1.4 beta5c:
...
  +/=  (0.62)   Depth: 20/32   00:00:13  40940kN
28...Kd7 29.Kg3 Ke8 30.Kxg4 Kd8 31.Kf3 Kd7 32.Ke4 Kc6 33.Kf4 Kd7 34.Kg3 Ke8 35.Kg2 Kf7 36.Kf3 Ke8 37.Ke3 Ke7 38.f4 Kd7 
  +/=  (0.62)   Depth: 21/32   00:00:23  70620kN
28...Kd7 29.Kg3 Ke8 30.Kxg4 Kd8 31.Kf3 Kd7 32.Ke4 Kc6 33.Ke3 Kd7 34.Ke2 Ke8 35.Kd3 Ke7 36.Ke4 Kd7 37.Kf3 Ke7 38.Kg3 Kd7 39.Kf4 Ke7 
  +/=  (0.62)   Depth: 22/35   00:00:42  127mN
28...Kd7 29.Kg3 Ke8 30.Kxg4 Kd8 31.Kf3 Kd7 32.Ke4 Kc6 33.Kf4 Kd7 34.Kg3 Kc6 35.Kg2 Kb7 36.Kf3 Kc6 37.Ke4 Kd7 38.Ke3 Ke7 39.f4 Kd7 
  +/=  (0.62)   Depth: 23/35   00:01:15  224mN

(,  08.03.2008)

And so on.

Comp Deep Fritz - Kramnik,V, Bahrain Brains m
[d]4k3/2p5/1p2p1p1/p1p1P1P1/P1P3p1/1P6/5PK1/8 b - - 0 1

Code: Select all

Analysis by Fritz 10:
...
  +/-  (0.71)   Depth: 17/31   00:00:01  1180kN
28...c6 29.Kg3 Kd7 30.Kxg4 Kc8 31.Kf4 Kb8 32.Kf3 Ka8 33.Ke3 Kb7 
  +/-  (0.71)   Depth: 18/40   00:00:01  2309kN
28...c6 29.Kg3 Kd7 30.Kxg4 Kc8 31.Kf4 Kb8 32.Kf3 Kc8 33.Ke2 Kc7 34.Kd2 
  +/-  (0.71)   Depth: 19/43   00:00:04  5233kN
28...c6 29.Kg3 Kd7 30.Kxg4 Kc8 31.Kf4 Kb8 32.Kf3 Kc8 33.Ke2 Kd7 34.f4 Ke7 35.Ke3 Kd7 36.Ke4 
  +/-  (0.71)   Depth: 20/44   00:00:07  9413kN
28...c6 29.Kg3 Kd7 30.Kxg4 Kc8 31.Kf4 Kd7 32.Ke3 Kc7 33.Kd3 Kb7 34.f4 Kc7 35.Ke3 Kd7 36.Ke2 
  +/-  (0.71)   Depth: 21/45   00:00:16  22728kN

(,  08.03.2008)

Fruit is also reporting +=0.62.
I doubt that a QUAD would help btw.


Game 2:

[Event "Bahrain Brains m"]
[Site "Manama"]
[Date "2002.10.04"]
[Round "2"]
[White "Kramnik, Vladimir"]
[Black "Comp Deep Fritz"]
[Result "1-0"]
[ECO "D27"]
[WhiteElo "2807"]
[PlyCount "113"]

1. d4 d5 2. c4 dxc4 3. Nf3 Nf6 4. e3 e6 5. Bxc4 c5 6. O-O a6 7. dxc5 Qxd1 8.
Rxd1 Bxc5 9. Kf1 b5 10. Be2 Bb7 11. Nbd2 Nbd7 12. Nb3 Bf8 13. a4 b4 14. Nfd2
Bd5 15. f3 Bd6 16. g3 e5 17. e4 Be6 18. Nc4 Bc7 19. Be3 a5 20. Nc5 Nxc5 21.
Bxc5 Nd7 22. Nd6+ Kf8 23. Bf2 Bxd6 24. Rxd6 Ke7 25. Rad1 Rhc8 26. Bb5 Nc5 27.
Bc6 Bc4+ 28. Ke1 Nd3+ 29. R1xd3 Bxd3 30. Bc5 Bc4 31. Rd4+ Kf6 32. Rxc4 Rxc6 33.
Be7+ Kxe7 34. Rxc6 Kd7 35. Rc5 f6 36. Kd2 Kd6 37. Rd5+ Kc6 38. Kd3 g6 39. Kc4
g5 40. h3 h6 41. h4 gxh4 42. gxh4 Ra7 43. h5 Ra8 44. Rc5+ Kb6 45. Rb5+ Kc6 46.
Rd5 Kc7 47. Kb5 b3 48. Rd3 Ra7 49. Rxb3 Rb7+ 50. Kc4 Ra7 51. Rb5 Ra8 52. Kd5
Ra6 53. Rc5+ Kd7 54. b3 Rd6+ 55. Kc4 Rd4+ 56. Kc3 Rd1 57. Rd5+ 1-0



[d]r7/4kppp/2R5/p3p3/Pp2P3/5PP1/1P5P/4K3 b - - 0 34
Fritz 10 says this is +=0.35 at depth 18. Kramnik said about this position that it is nearly lost for black

Now, a couple of fine examples of waiting-and torturing strategy by Kramnik:

[d]r7/6pp/3k1p2/p1R1p3/Pp2P3/5PP1/1P1K3P/8 w - - 0 37

Rd5+! Computer prefers here Rb5. Kramnik plays Rd5+: he can go Rb5 whenever he wants, but there is difference whether you have your King on d6 or in c6(e6).

Now we are here:
[d]r7/7p/2k2p2/p2Rp1p1/PpK1P3/5PP1/1P5P/8 w - - 0 40
My Fritz 10 already realizes that black is bad, but as white it would never see the move 40. h3! The point is black cannot do anything, and h3 improves, waits, doesn't miss or spoil anything...its superior to rushing out with Kd3 as my Fritz suggests

This one is my favorite in this game.
[d]8/r7/2k2p1p/p2Rp3/PpK1P2P/5P2/1P6/8 w - - 0 43
Kramnik plays 43 h5!- a move that all commentators said is clearly superior to Rd8 that comps are suggesting with huge evals. My Fritz is quite happy with Rd8 with a +1.5 score.

Kramnik,V - Comp Deep Fritz, Bahrain Brains m
[d]8/r7/2k2p1p/p2Rp3/PpK1P2P/5P2/1P6/8 w - - 0 1

Code: Select all

Analysis by Fritz 10:
...
  +-  (1.54)   Depth: 19/46   00:00:38  68024kN
43.Rd8 Rc7 44.h5 f5 45.Rd5 Kb6+ 46.Kb3 fxe4 47.fxe4 Re7 48.Rb5+ Ka6 49.Kc4 Re8 50.Rd5 Re6 51.Rc5 
  +-  (1.48)   Depth: 20/65   00:01:19  139mN

(,  08.03.2008)

Fruit suggests a pointless move-Kd3 here.
Toga likes Rd8, but with a lesser score than Fritz.

This game is ruthless. If you replay it, you can see that after some point Fritz did not have a single counter-chance, a point where it could breathe. Play any alternative line suggested by computer-you will never get a position as cristal-winning as what Kramnik got.[/quote]

I strongly disagree with you.
There are positions in endgame that GM are better but there are also positions in endgame that computers are better.

You cannot use wrong evaluation for this discussion and you have to show wrong moves.

Your claim that computers cannot find kramnik's move Rd5+ is clearly wrong

Kramnik,V - Comp Deep Fritz, Bahrain Brains m 2002
r7/6pp/3k1p2/p1R1p3/Pp2P3/5PP1/1P1K3P/8 w - - 0 1

Analysis by Rybka 2.3.2a 32-bit :

37.Rc5-d5+ Kd6-e6 38.Rd5-b5
= (0.20) Depth: 5 00:00:18
37.Rc5-d5+ Kd6-e6 38.Rd5-b5 Ra8-a7
² (0.26) Depth: 6 00:00:18 1kN
37.Rc5-d5+ Kd6-c6 38.Kd2-e3 g7-g6 39.b2-b3
² (0.28) Depth: 7 00:00:18 5kN
37.Rc5-d5+ Kd6-c6 38.Kd2-e3 g7-g6 39.b2-b3 Ra8-a7
² (0.27) Depth: 8 00:00:18 6kN
37.Rc5-d5+ Kd6-c6 38.Kd2-e3 g7-g6 39.b2-b3 Ra8-a7 40.Rd5-b5
² (0.26) Depth: 9 00:00:18 9kN
37.Rc5-d5+ Kd6-e6 38.Rd5-b5 Ra8-a7 39.Kd2-e3 h7-h6 40.Rb5-b6+ Ke6-d7 41.h2-h3
² (0.26) Depth: 10 00:00:18 34kN
37.Rc5-b5 Ra8-a7 38.Kd2-e3 Kd6-c6 39.h2-h4 g7-g6 40.Rb5-d5
² (0.27) Depth: 10 00:00:19 42kN
37.Rc5-b5 Kd6-c7 38.Kd2-e3 g7-g6 39.Rb5-d5 Kc7-c6 40.b2-b3 Ra8-a7 41.Rd5-b5 h7-h6
² (0.27) Depth: 11 00:00:19 59kN
37.Rc5-d5+ Kd6-e6 38.Rd5-b5 Ra8-a7 39.Kd2-e3 h7-h6 40.Rb5-b6+ Ke6-d7 41.h2-h4 Kd7-c7
² (0.28) Depth: 11 00:00:19 78kN
37.Rc5-d5+ Kd6-e6 38.Rd5-b5 Ra8-a7 39.Kd2-e3 h7-h6 40.Rb5-b6+ Ke6-d7 41.h2-h4 Kd7-c7 42.Rb6-b5
² (0.28) Depth: 12 00:00:20 88kN
37.Rc5-d5+ Kd6-e6 38.Rd5-b5 Ra8-a7 39.Kd2-e3 h7-h6 40.Rb5-b6+ Ke6-d7 41.h2-h4 Kd7-c7 42.Rb6-b5 g7-g6
² (0.28) Depth: 13 00:00:21 111kN
37.Rc5-d5+ Kd6-e6 38.Rd5-b5 g7-g5 39.Rb5-b6+ Ke6-f7 40.Rb6-b7+ Kf7-g6 41.Rb7-c7 h7-h5 42.Rc7-c5 Kg6-f7 43.Kd2-e3 Kf7-e7
² (0.28) Depth: 14 00:00:21 289kN
37.Rc5-d5+ Kd6-e6 38.Kd2-e3 g7-g6 39.f3-f4 Ra8-a7 40.f4xe5 f6xe5 41.h2-h3 h7-h5 42.Rd5-b5 Ra7-a6 43.Rb5-c5 Ke6-d6
² (0.29) Depth: 15 00:00:26 640kN
37.Rc5-d5+ Kd6-e6 38.Kd2-d3 h7-h5 39.Kd3-c4 g7-g5 40.h2-h3 f6-f5 41.Rd5-c5 f5xe4 42.f3xe4 Ke6-d6 43.Rc5-d5+ Kd6-e6
² (0.53) Depth: 16 00:00:37 1404kN
37.Rc5-d5+ Kd6-c6 38.Kd2-d3 g7-g6 39.Kd3-c4 g6-g5 40.Kc4-d3 Kc6-c7 41.Kd3-e3 Ra8-a6 42.h2-h4 g5xh4 43.g3xh4 Kc7-c6
² (0.53) Depth: 17 00:00:40 1613kN
37.Rc5-d5+ Kd6-c6 38.Kd2-e3 Kc6-b6 39.Rd5-d7 Ra8-a7 40.Rd7xa7 Kb6xa7 41.Ke3-d3 Ka7-b7 42.h2-h4 Kb7-b6 43.Kd3-c4 Kb6-c6
² (0.53) Depth: 18 00:00:59 2975kN
37.Rc5-d5+ Kd6-c7 38.Kd2-e3 Kc7-b6 39.Rd5-d7 Ra8-a7 40.Rd7xa7 Kb6xa7 41.Ke3-d3 Ka7-b7 42.h2-h4 Kb7-c6 43.Kd3-c4 g7-g6
² (0.52) Depth: 19 00:01:44 6092kN

(, 09.03.2008)

Rbka also suggests h4 similiar idea to h3 and kramnik played h4 later in the game

Kramnik,V - Comp Deep Fritz, Bahrain Brains m 2002
r7/7p/2k2p2/p2Rp1p1/PpK1P3/5PP1/1P5P/8 w - - 0 1

Analysis by Rybka 2.3.2a 32-bit :

40.Kc4-d3 Kc6-c7 41.Kd3-e3 Kc7-c6 42.f3-f4 e5xf4+ 43.g3xf4 h7-h6 44.Rd5-f5 g5xf4+ 45.Ke3xf4
² (0.50) Depth: 15 00:00:00 19kN
40.Kc4-d3 Kc6-c7 41.Kd3-e3 Kc7-c6 42.f3-f4 e5xf4+ 43.g3xf4 h7-h6 44.Rd5-f5 g5xf4+ 45.Ke3xf4
² (0.50) Depth: 16 00:00:00 35kN
40.Kc4-d3 Kc6-c7 41.Kd3-e3 Kc7-c6 42.f3-f4 e5xf4+ 43.g3xf4 h7-h6 44.Rd5-f5 g5xf4+ 45.Ke3xf4
² (0.50) Depth: 17 00:00:01 61kN
40.Kc4-d3 Kc6-c7 41.Kd3-e3 Kc7-c6 42.f3-f4 e5xf4+ 43.g3xf4 h7-h6 44.Rd5-f5 g5xf4+ 45.Ke3xf4
² (0.50) Depth: 18 00:00:01 111kN
40.Kc4-d3 Kc6-c7 41.Kd3-e3 Kc7-c6 42.f3-f4 e5xf4+ 43.g3xf4 g5xf4+ 44.Ke3xf4 Ra8-g8 45.Rd5xa5 Rg8-g2 46.b2-b3 Rg2-b2
² (0.53) Depth: 19 00:00:08 625kN
40.h2-h4 g5xh4 41.g3xh4 h7-h6 42.Kc4-d3 Ra8-a7 43.Kd3-e3 Kc6-b6 44.f3-f4 e5xf4+ 45.Ke3xf4 Ra7-g7 46.Rd5-b5+ Kb6-c6
± (0.88) Depth: 19 00:00:31 2173kN
40.h2-h4 g5xh4 41.g3xh4 h7-h6 42.Kc4-d3 Ra8-a7 43.Kd3-e3 Kc6-b6 44.f3-f4 e5xf4+ 45.Ke3xf4 Ra7-g7 46.Rd5-b5+ Kb6-c6
± (0.88) Depth: 20 00:00:35 2418kN
40.h2-h4 g5xh4 41.g3xh4 h7-h6 42.Kc4-d3 Ra8-a7 43.b2-b3 Ra7-a6 44.Kd3-e3 Kc6-c7 45.f3-f4 e5xf4+ 46.Ke3xf4 Ra6-c6
± (0.88) Depth: 21 00:00:47 3206kN

(, 09.03.2008)

[Uri Blass]

I still think that a player like Kramnik has quite fair chances against any engine in a classical match. Despite that huge rating difference between todays and 5-year-old engines, the pattern is the same: the engines are much worse on strategy and enourmously better in tactics. Engines are likely to lose many non-tactical endgames that a human should be able to hold. Rybka is notable in this context: without tablebases I think its one of the most clueless in endgame.

I disagree.
Public rybka has some bugs that it evaluate unstoppable passed pawn as more than a queen but this bug was probably fixed.

I do not think that rybka is weaker than other engines in endgames and I also do not think that engines are likely to lose many non tactical endgames.

Uri

Then you should analyze super-GM endgames with your favorite engine
Recent example was Rybka evaluating an endgame with a white-colored white bishop, white and black pawns on h-file and black king on h8 as +-4.50-or-something like that. Tablebases might help in this particular case, but you know better than me that they are not a solution-you get a huge speed decrease because of them and it doesn't make real difference in average.

Kramnik-Fritz games from Bahrain match are a thrilling piece of example of misplayed endgames by an engine. I don't think that there has been a remarkable advance in that aspect of play since then.

I know that Rybka misevaluate some endgames but humans also do mistakes in other endgame that rybka does not do.

I cannot get the conclusion that rybka is no super-GM in endgames
based only on the positions that she evaluates wrong.

Tablebases are minority of the endgames and they are not very important.
Rybka won against benjamin also because of playing better endgames than benjamin.

Uri

[Dirt]

Anyway, after this draw against a 2542 (although he was top-20 once), I think Rybka could try a 'normal' match against Nakamura who is close to 2700... at least without material odds. Maybe a Chess960 match would be good, to avoid long theory variations. In such a match, we want to see as many Rybka engine moves as possible, no theory battles from memory or book. (Rybka is Chess960 computer world champion, too.)

Nakamura getting draw odds and always playing white might make a good match, if there is a long enough time control. I think he might be relatively poor under these conditions for someone of his rating, but it might also help him shore up his defensive abilities.

Playing Fischer Random against a computer is a good way to lose faster and with more certainty. I think FRC may be a better game in some ways for top chess players, but spectators don't want watch what they don't play; and the reasons to switch from regular chess don't really come into play for those who aren't GMs.

[playjunior]

Uri, you indeed might find one engine that finds one of the moves that Kramnik made...especially if you look at the PV-it is clear Rybka missed the whole point there...

The whole point of playing h3 and then only h4 is that white wants to pass a move. He's better, black is cramped, does not have a clear plan, lets give him a move and see what does he do.

My point is that the difference between good and great is whether you play h2-h4 at once or h2-h3 and then h3-h4.

You can take any Kramnik endgame victory against an 2600+ player and you will almost for sure find a couple of finest moves that a computer does not see. However, I strongly doubt that you can find clear wins(with a computer) in endgames where he is better but does not win.

You can start with his recent games in Catalan (say, from the last Tal memorial in Moscow).