Houdini-Rybka (150m + 30s) LIVE! - TCEC Elite Match Season 1

[Martin Thoresen]

Thanks, Matthias.

For those that is wondering what is happening during game 14, well:
The webhost is upgrading their storage systems and this started around 00:00 CET and is expected to last 2-3 hours.

Unfortunately there's nothing I can do about that.

In any case, 2 hours have passed now so you can expect the website to be back very soon.

Best,
Martin

[BB+]

Just to wrap up the "Rybka poor time-management" story, Houdini managed to win that game.[Round "12"][White "Houdini"][Black "Rybka"][Result "1-0"]
1. e4 c6 2. d4 d5 3. e5 Bf5 4. Nf3 e6 5. Be2 Ne7 6. O-O c5 7. Nc3 Nec6 8. Bg5 Qa5 9. a3 Nxd4 10. Nxd4 cxd4 11. Nb5 h6 12. b4 Qb6 13. Bc1 Nc6 14. Bb2 Bg4 15. Bxg4 Qxb5 16. f4 g6 17. Be2 Qb6 18. Kh1 a6 19. Rb1 Be7 20. Rf3 Rc8 21. Rh3 Qa7 22. Qf1 b5 23. Re1 Kf8 24. Bd3 Kg7 25. Qf2 Rcf8 26. Rf3 Bd8 27. Rf1 Bb6 28. Qg3 Qb8 29. Qh4 Qd8 30. Qh3 h5 31. Bc1 Bc7 32. Qg3 Qd7 33. Bb2 h4 34. Qg5 Rh5 35. Qg4 Bd8 36. Rc1 Bb6 37. Rh3 Qe7 38. Rb1 Rhh8 39. Rd1 Rc8 40. Rf1 Rh5 41. Qe2 Rh7 42. Qg4 Rch8 43. Ra1 Qd7 44. Qe2 Bd8 45. Bc1 Be7 46. Bd2 Re8 47. Be1 Rc8 48. Bf2 Rhh8 49. Qg4 Qd8 50. Rf3 Rh5 51. Re1 Rc7 52. Bg1 Qh8 53. h3 Qc8 54. Rff1 Qd7 55. Rc1 Rc8 56. Ra1 Bd8 57. Rfc1 Qb7 58. Bf2 Rh8 59. Rcb1 Qe7 60. Qe2 Rb8 61. Be1 Rh5 62. Qg4 Rh6 63. Rb3 Rh8 64. Bf2 Rc8 65. Rbb1 Rh5 66. Be1 Rb8 67. Ra2 Rh8 68. Bf2 Bc7 69. Raa1 Rh5 70. Rc1 Bb6 71. Rg1 Rhh8 72. Rgb1 Rb7 73. Rd1 Rh6 74. Rdc1 Rh5 75. Be1 Rb8 76. Ra2 Rhh8 77. Bd2 Qd7 78. Rca1 Bd8 79. Be1 Qc7 80. Rb2 Qa7 81. Bf2 Rc8 82. Rf1 Qe7 83. Kg1 Bc7 84. Rbb1 Rh6 85. Rfe1 Rh5 86. Kh1 Bb6 87. Rf1 Rhh8 88. Ra1 Rh6 89. Rfb1 Rch8 90. Be1 Rb8 91. Qe2 Bd8 92. Rc1 Qb7 93. Qg4 Rc8 94. Rab1 Rh8 95. Bf2 Qd7 96. Rd1 Qc7 97. Rf1 Be7 98. Bg1 Rh6 99. Rbc1 Rh5 100. c3 dxc3 101. Rxc3 Qd7 102. Rc2 Rhh8 103. Bb6 Qe8 104. Be3 Qd7 105. Kg1 Rh5 106. Rfc1 Rhh8 107. Kh2 Kh7 108. Rf1 Kg7 109. Bg1 Rhf8 110. Re2 Rcd8 111. Kh1 d4 112. Bh2 a5 113. bxa5 b4 114. Ref2 Qd5 115. f5 exf5 116. Rxf5 Kg8 117. Rh5 Qe6 118. Bf5 Nxe5 119. Bxe5 Qxe5 1-0As one commenter put it, top-level chess doesn't start until move 100 these days. There's currently a web-hosting break, so I can't look at the full data from the TCEC website. My recollection is that Rybka was down to increments (30 seconds) more-or-less, while Houdini had maybe 5 extra minutes. It looks like the f5-push just became more-and-more feasible, and Black cracked (moves 112/113 looks weak, but maybe nothing really improves).

Incidentally, what might(?) have happened with move ~100 is that Houdini cleared its hash table when the 50-move rule was on the horizon (somewhere RH confirmed that it did this, but I can't track it down), and then on re-search it didn't get as deep, and thought it was slightly winning rather than slightly losing.

[Jeremy Bernstein]

Just to wrap up the "Rybka poor time-management" story, Houdini managed to win that game.[Round "12"][White "Houdini"][Black "Rybka"][Result "1-0"]
1. e4 c6 2. d4 d5 3. e5 Bf5 4. Nf3 e6 5. Be2 Ne7 6. O-O c5 7. Nc3 Nec6 8. Bg5 Qa5 9. a3 Nxd4 10. Nxd4 cxd4 11. Nb5 h6 12. b4 Qb6 13. Bc1 Nc6 14. Bb2 Bg4 15. Bxg4 Qxb5 16. f4 g6 17. Be2 Qb6 18. Kh1 a6 19. Rb1 Be7 20. Rf3 Rc8 21. Rh3 Qa7 22. Qf1 b5 23. Re1 Kf8 24. Bd3 Kg7 25. Qf2 Rcf8 26. Rf3 Bd8 27. Rf1 Bb6 28. Qg3 Qb8 29. Qh4 Qd8 30. Qh3 h5 31. Bc1 Bc7 32. Qg3 Qd7 33. Bb2 h4 34. Qg5 Rh5 35. Qg4 Bd8 36. Rc1 Bb6 37. Rh3 Qe7 38. Rb1 Rhh8 39. Rd1 Rc8 40. Rf1 Rh5 41. Qe2 Rh7 42. Qg4 Rch8 43. Ra1 Qd7 44. Qe2 Bd8 45. Bc1 Be7 46. Bd2 Re8 47. Be1 Rc8 48. Bf2 Rhh8 49. Qg4 Qd8 50. Rf3 Rh5 51. Re1 Rc7 52. Bg1 Qh8 53. h3 Qc8 54. Rff1 Qd7 55. Rc1 Rc8 56. Ra1 Bd8 57. Rfc1 Qb7 58. Bf2 Rh8 59. Rcb1 Qe7 60. Qe2 Rb8 61. Be1 Rh5 62. Qg4 Rh6 63. Rb3 Rh8 64. Bf2 Rc8 65. Rbb1 Rh5 66. Be1 Rb8 67. Ra2 Rh8 68. Bf2 Bc7 69. Raa1 Rh5 70. Rc1 Bb6 71. Rg1 Rhh8 72. Rgb1 Rb7 73. Rd1 Rh6 74. Rdc1 Rh5 75. Be1 Rb8 76. Ra2 Rhh8 77. Bd2 Qd7 78. Rca1 Bd8 79. Be1 Qc7 80. Rb2 Qa7 81. Bf2 Rc8 82. Rf1 Qe7 83. Kg1 Bc7 84. Rbb1 Rh6 85. Rfe1 Rh5 86. Kh1 Bb6 87. Rf1 Rhh8 88. Ra1 Rh6 89. Rfb1 Rch8 90. Be1 Rb8 91. Qe2 Bd8 92. Rc1 Qb7 93. Qg4 Rc8 94. Rab1 Rh8 95. Bf2 Qd7 96. Rd1 Qc7 97. Rf1 Be7 98. Bg1 Rh6 99. Rbc1 Rh5 100. c3 dxc3 101. Rxc3 Qd7 102. Rc2 Rhh8 103. Bb6 Qe8 104. Be3 Qd7 105. Kg1 Rh5 106. Rfc1 Rhh8 107. Kh2 Kh7 108. Rf1 Kg7 109. Bg1 Rhf8 110. Re2 Rcd8 111. Kh1 d4 112. Bh2 a5 113. bxa5 b4 114. Ref2 Qd5 115. f5 exf5 116. Rxf5 Kg8 117. Rh5 Qe6 118. Bf5 Nxe5 119. Bxe5 Qxe5 1-0As one commenter put it, top-level chess doesn't start until move 100 these days. There's currently a web-hosting break, so I can't look at the full data from the TCEC website. My recollection is that Rybka was down to increments (30 seconds) more-or-less, while Houdini had maybe 5 extra minutes. It looks like the f5-push just became more-and-more feasible, and Black cracked (moves 112/113 looks weak, but maybe nothing really improves).

Incidentally, what might(?) have happened with move ~100 is that Houdini cleared its hash table when the 50-move rule was on the horizon (somewhere RH confirmed that it did this, but I can't track it down), and then on re-search it didn't get as deep, and thought it was slightly winning rather than slightly losing.

Rybka 4 loves ...Qd5 for a very long time on my SP 32-bit XP VM. Only after 5:30, ply 14 in rybkatalk (so 17 in the real world, I guess), does it move on to ...Kh8. I'll wait and see what it decides after 20 minutes or so.

[Jeremy Bernstein]

Just to wrap up the "Rybka poor time-management" story, Houdini managed to win that game.[Round "12"][White "Houdini"][Black "Rybka"][Result "1-0"]
1. e4 c6 2. d4 d5 3. e5 Bf5 4. Nf3 e6 5. Be2 Ne7 6. O-O c5 7. Nc3 Nec6 8. Bg5 Qa5 9. a3 Nxd4 10. Nxd4 cxd4 11. Nb5 h6 12. b4 Qb6 13. Bc1 Nc6 14. Bb2 Bg4 15. Bxg4 Qxb5 16. f4 g6 17. Be2 Qb6 18. Kh1 a6 19. Rb1 Be7 20. Rf3 Rc8 21. Rh3 Qa7 22. Qf1 b5 23. Re1 Kf8 24. Bd3 Kg7 25. Qf2 Rcf8 26. Rf3 Bd8 27. Rf1 Bb6 28. Qg3 Qb8 29. Qh4 Qd8 30. Qh3 h5 31. Bc1 Bc7 32. Qg3 Qd7 33. Bb2 h4 34. Qg5 Rh5 35. Qg4 Bd8 36. Rc1 Bb6 37. Rh3 Qe7 38. Rb1 Rhh8 39. Rd1 Rc8 40. Rf1 Rh5 41. Qe2 Rh7 42. Qg4 Rch8 43. Ra1 Qd7 44. Qe2 Bd8 45. Bc1 Be7 46. Bd2 Re8 47. Be1 Rc8 48. Bf2 Rhh8 49. Qg4 Qd8 50. Rf3 Rh5 51. Re1 Rc7 52. Bg1 Qh8 53. h3 Qc8 54. Rff1 Qd7 55. Rc1 Rc8 56. Ra1 Bd8 57. Rfc1 Qb7 58. Bf2 Rh8 59. Rcb1 Qe7 60. Qe2 Rb8 61. Be1 Rh5 62. Qg4 Rh6 63. Rb3 Rh8 64. Bf2 Rc8 65. Rbb1 Rh5 66. Be1 Rb8 67. Ra2 Rh8 68. Bf2 Bc7 69. Raa1 Rh5 70. Rc1 Bb6 71. Rg1 Rhh8 72. Rgb1 Rb7 73. Rd1 Rh6 74. Rdc1 Rh5 75. Be1 Rb8 76. Ra2 Rhh8 77. Bd2 Qd7 78. Rca1 Bd8 79. Be1 Qc7 80. Rb2 Qa7 81. Bf2 Rc8 82. Rf1 Qe7 83. Kg1 Bc7 84. Rbb1 Rh6 85. Rfe1 Rh5 86. Kh1 Bb6 87. Rf1 Rhh8 88. Ra1 Rh6 89. Rfb1 Rch8 90. Be1 Rb8 91. Qe2 Bd8 92. Rc1 Qb7 93. Qg4 Rc8 94. Rab1 Rh8 95. Bf2 Qd7 96. Rd1 Qc7 97. Rf1 Be7 98. Bg1 Rh6 99. Rbc1 Rh5 100. c3 dxc3 101. Rxc3 Qd7 102. Rc2 Rhh8 103. Bb6 Qe8 104. Be3 Qd7 105. Kg1 Rh5 106. Rfc1 Rhh8 107. Kh2 Kh7 108. Rf1 Kg7 109. Bg1 Rhf8 110. Re2 Rcd8 111. Kh1 d4 112. Bh2 a5 113. bxa5 b4 114. Ref2 Qd5 115. f5 exf5 116. Rxf5 Kg8 117. Rh5 Qe6 118. Bf5 Nxe5 119. Bxe5 Qxe5 1-0As one commenter put it, top-level chess doesn't start until move 100 these days. There's currently a web-hosting break, so I can't look at the full data from the TCEC website. My recollection is that Rybka was down to increments (30 seconds) more-or-less, while Houdini had maybe 5 extra minutes. It looks like the f5-push just became more-and-more feasible, and Black cracked (moves 112/113 looks weak, but maybe nothing really improves).

Incidentally, what might(?) have happened with move ~100 is that Houdini cleared its hash table when the 50-move rule was on the horizon (somewhere RH confirmed that it did this, but I can't track it down), and then on re-search it didn't get as deep, and thought it was slightly winning rather than slightly losing.

Rybka 4 loves ...Qd5 for a very long time on my SP 32-bit XP VM. Only after 5:30, ply 14 in rybkatalk (so 17 in the real world, I guess), does it move on to ...Kh8. I'll wait and see what it decides after 20 minutes or so.

After 21:54 (ply 17), we're at ...f5. Anyhow, anyone accusing Martin of rigging the games, or blaming the UI for strange behavior, should run some simple tests. This is just how R4 works in this position.

Jeremy

[Martin Thoresen]

Yeah well I don't know what to say really, some people throw many strange accusations in the air.

Thanks for testing out the position Jeremy. I can't provide anything else than the logs (which I posted in talkchess).

Best,
Martin

[BB+]

I might say that, for the first long shifting part of this game 12, Rybka seemed to outplay Houdini (slightly), but it was only enough to move the eval from equal to -0.2 or so. I've never liked these French (or Caro-Kann here) games where Black gets cxd4 in and White chooses not to re-capture. Of course, there is often little negative from White's view (the pawn can usually be won back, and if not, Black frequently has no entry point -- plus White here got the bishop pair, which again makes Black reluctant to open the position).

[BB+]

r5k1/3b1pb1/2p2pp1/p7/4RBP1/q1P5/B3QP2/6K1 w - - 0 34Side puzzle for humans (Game 16): Black just played 33...Qxa3 (Rybka, of course, did not) -- how do you refute this?
Side puzzle for centaurs/analysts: does White indeed win the resulting Q vs BB endgame (2 or 3 pawns per side, and maybe with rooks)? I fully expect "yes", but it seems to hover around +4 for a little too long for my taste.

[BB+]

Game 16 has now become RBB+2 vs RB+5. There is talk of Rybka "over-valuing" passed pawns (again?), though it might be a more general problem of pieces versus pawns. Certainly in the opening a piece is worth almost 4 pawns, but in the endgame, while the "standard" value is 3, this can vary significantly depending upon if the pawns are weak/dangerous.
r7/5pk1/2p2p2/6p1/p1B2B2/b1P3R1/5PK1/8 w - - 2 41It should be an interesting endgame if nothing else, though I expect a draw in the end.

[BB+]

Game 16, Move 54, Rybka did not play what I think is a clever move here (I get 0.00 scores with TBs):
8/5p2/r1pR1p2/2P1k1p1/8/p4P2/B4K2/8 b - - 2 54Black can play 54...Rb6, and I think the draw is easier than after 54...f5. The main point is that after 55. Rd5+ cxd5 56. cxb6 Kd6 57. b7 Kc7 58. Bxd5 White makes no progress. If the White King goes to win the a3-pawn, Black advances his kingside pawns to trade off White's f-pawn.

Still looks to be headed for a draw, though Stockfish gets excited a lot.

[BB+]

Game 28 is another quite interesting game.
[Round "28"][White "Rybka"][Black "Houdini"][Result "*"]1. d4 Nf6 2. c4 c5 3. Nf3 cxd4 4. Nxd4 Nc6 5. g3 d6 6. Bg2 Bd7 7. Nc3 g6 8. O-O Bg7 9. Nc2 Ng4 10. Bd2 h5 11. h3 Nh6 12. Rc1 Qc8 13. Kh2 Rb8 14. f4 f5 15. Ne1 h4 16. gxh4 Nf7 17. Rg1 Rxh4 18. Bxc6 bxc6 19. Rxg6 Rh7 20. c5 Kf8 21. b3 Nh6 22. Kg2 d5 23. Nf3 Be8 24. Rg3 d4 25. Na4 Bh5 26. Nb2 Bxf3+ 27. exf3 Bf6 28. Nd3 Bh4 29. Rg6 Qe8 30. Ne5 Nf7 31. Nxc6 Nd8 32. Ne5 Nf7 33. Ng4 fxg4 34. hxg4 Bf6 35. Qc2 d3 36. Qc4It doesn't seem as though the h4-push went as well as Black had hoped, and White had a definite advantage in the mid-20s of the move count. There were some nice lines after 28. Qe2 (Black can then "trap" the g3 Rook via 28... Bh4 29. Rg6 Kf7 30. Rg5, but White's attack rolls through if Black wins the exchange), while the game line was also cute with the Knights hopping. White gets nothing from 33. Nd3 Nh8 (though lines with 34. Rg5 can be interesting for humans -- note that if the Rook leaves the g-file (Ra6 or Re6), Black should draw after Rg7+), so Rybka has tried the "piece sac" of 33. Ng4 instead. After 34... Bf6, Black is aiming to round-up the Rook via Nh8 (White can't play 36. Qxd3 because of 36... Rd8, when either the d2-bishop hangs, or the Rg6 goes -- though there are some tactics here after 37. Qc2 Nh8 38. Rxf6+ because Black's Rh7 is also unguarded).

As a sidenote, it seems that Houdini typically prefers a piece to 3 pawns, while Rybka (at least relatively) prefers the pawns. A number of recent games have had this motif [and the current evals reflect this too, Rybka says +0.96 and Houdini +0.15].