Latest SF dev beats H6 Pro in the Goldin vs Korzubov endgame

Discussion of computer chess matches and engine tournaments.

Moderators: bob, hgm, Harvey Williamson

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
User avatar
Nordlandia
Posts: 2476
Joined: Fri Sep 25, 2015 7:38 pm
Location: Sortland, Norway

Latest SF dev beats H6 Pro in the Goldin vs Korzubov endgame

Post by Nordlandia » Wed Feb 14, 2018 8:21 pm

Few months ago this was considered engine draw.

[d]6k1/pp3p2/6p1/3p4/2pN4/P1P2P1P/1P1qR1P1/6K1 b - - 0 1

TC: 60min+45s
Hash: 2048 Mb allocated per engine
GUI: CuteChess 1.0.0

[pgn][Event "?"]
[Site "Q9650 | 4-Core | Probing 5-men | 6-men syzygy adjudication"]
[Date "2018.02.14"]
[Round "?"]
[White "Houdini 6"]
[Black "stockfish_x64"]
[Result "0-1"]
[FEN "6k1/pp3p2/6p1/3p4/2pN4/P1P2P1P/1P1qR1P1/6K1 b - - 0 1"]
[GameEndTime "2018-02-14T21:04:42.701 W. Europe Standard Time"]
[GameStartTime "2018-02-14T18:43:10.316 W. Europe Standard Time"]
[PlyCount "71"]
[SetUp "1"]
[Termination "adjudication"]
[TimeControl "3600+30"]

1... Qc1+ {+1.61/44 141s} 2. Kh2 {-1.48/34 213s} g5 {+1.65/47 219s}
3. g3 {-1.21/33 101s} Qf1 {+1.80/41 32s} 4. h4 {-1.52/35 144s}
gxh4 {+1.79/44 48s} 5. gxh4 {-1.52/38 80s} a6 {+1.95/47 171s}
6. Kg3 {-1.52/35 18s} Kg7 {+1.95/50 118s} 7. Rh2 {-1.52/36 75s}
b5 {+2.04/44 34s} 8. Re2 {-1.52/36 80s} Kf6 {+2.12/44 45s} 9. Rg2 {-1.52/36 17s}
Qh1 {+2.28/42 38s} 10. Re2 {-1.60/38 204s} Qg1+ {+2.34/48 122s}
11. Rg2 {-1.62/36 109s} Qf1 {+2.49/45 188s} 12. Rh2 {-1.83/36 155s}
Qe1+ {+2.52/43 72s} 13. Kg4 {-1.90/34 127s} Qg1+ {+2.60/44 175s}
14. Kh3 {-1.90/36 129s} Qe3 {+2.63/42 37s} 15. Kg4 {-1.91/31 27s}
Qe8 {+2.63/45 63s} 16. f4 {-1.91/35 116s} Qg8+ {+2.86/42 257s}
17. Kh3 {-2.01/35 126s} Qg6 {+3.02/42 37s} 18. Rd2 {-2.10/30 17s}
Qg1 {+3.02/46 53s} 19. Rh2 {-1.99/32 108s} Qe3+ {+3.09/41 39s}
20. Kg4 {-1.62/37 62s} Qg1+ {+3.17/45 278s} 21. Kh3 {-0.01/99 1.2s}
Qd1 {+3.17/48 79s} 22. Rg2 {-2.18/36 252s} Qd3+ {+3.24/44 88s}
23. Kg4 {-2.42/37 190s} Qg6+ {+3.32/45 52s} 24. Kh3 {-2.42/37 26s}
Qe4 {+3.39/48 120s} 25. f5 {-2.42/38 194s} Qf4 {+3.47/47 119s}
26. Rg4 {-2.42/39 131s} Qc1 {+3.47/44 50s} 27. Rg2 {-2.42/41 100s}
Qf4 {+3.47/48 75s} 28. Rg4 {-0.01/74 65s} Qf2 {+3.47/47 57s}
29. Rg2 {-2.51/41 703s} Qf1 {+3.47/49 136s} 30. Kg3 {-2.79/34 159s}
Kg7 {+3.47/50 91s} 31. Re2 {-2.51/31 41s} a5 {+3.62/37 33s}
32. f6+ {-2.95/31 175s} Kg6 {+4.94/40 637s} 33. Rf2 {-3.14/31 342s}
Qd3+ {+5.17/31 29s} 34. Nf3 {-3.42/31 186s} b4 {+5.44/30 29s}
35. axb4 {-2.82/25 31s} axb4 {+5.44/35 73s} 36. cxb4 {-3.49/27 99s}
Kxf6 {+5.53/33 30s, Black wins by adjudication: user decision} 0-1

[/pgn]

tpoppins
Posts: 919
Joined: Tue Nov 24, 2015 8:11 pm
Location: upstate

Re: Latest SF dev beats H6 Pro in the Goldin vs Korzubov end

Post by tpoppins » Wed Feb 14, 2018 11:51 pm

Nordlandia wrote:Few months ago this was considered engine draw.

[d]6k1/pp3p2/6p1/3p4/2pN4/P1P2P1P/1P1qR1P1/6K1 b - - 0 1

TC: 60min+45s
Hash: 2048 Mb allocated per engine
GUI: CuteChess 1.0.0
*snip*
I'm surprised you still remember this position.

This is the original thread from last summer: Goldin vs Korzubov, Dushanbe 1980. You requested LTC games on strong hardware for this position and you got half a dozen ranging from 4h/game to 10h/game, by several engines, including Houdini Tactical and Komodo with null-move pruning off. The results were inconclusive.

When I saw your post above I thought for a second that you may have been analyzing this position on and off since then, and after hundreds of (CPU)hours of analysis finally discovered a definitive solution... And what do I find? A single game at TC 1/4th-1/10th of the games in that thread. As usual, you don't make even an attempt at independent analysis, it's just a computer-generated PGN. Why did White lose this game? I bet you don't have a clue about this position any more than you did last summer.

The very least you could have done (if you can't or won't analyze yourself) is at least make it a match and worthy of being posted in the Tournaments and Matches section. Say, a hundred games at this short TC. Then you could at least say with a certain degree of confidence "Black tends to win in xx% of cases" or "White seems to draw this or that proportion of games". True, that would have been replacing true knowledge with statistics (i.e. a very rough approximation of the truth) but that would have still been better than nothing at all (which is all the above game is worth, I'm afraid).

User avatar
Nordlandia
Posts: 2476
Joined: Fri Sep 25, 2015 7:38 pm
Location: Sortland, Norway

Re: Latest SF dev beats H6 Pro in the Goldin vs Korzubov end

Post by Nordlandia » Thu Feb 15, 2018 12:21 am

T. Poppins: you're right. I can't draw conculsions based on single games. Based on your reply I want to launch 100 game match between H6 and SF Dev in this position as well ->

[d]5k2/ppp2ppp/1b6/8/6b1/2P5/PP4PP/RN2K3 w - -

User avatar
Nordlandia
Posts: 2476
Joined: Fri Sep 25, 2015 7:38 pm
Location: Sortland, Norway

Re: Latest SF dev beats H6 Pro in the Goldin vs Korzubov end

Post by Nordlandia » Thu Feb 15, 2018 10:38 am

What about asking top human players about this endgame?

TC: 4H + 30s

[pgn][Event "?"]
[Site "Q9650 | 4-Core | Probing 5-men | 6-men syzygy adjudication"]
[Date "2018.02.14"]
[Round "?"]
[White "Houdini 6"]
[Black "stockfish_x64"]
[Result "1/2-1/2"]
[FEN "6k1/pp3p2/6p1/3p4/2pN4/P1P2P1P/1P1qR1P1/6K1 b - - 0 1"]
[GameEndTime "2018-02-15T08:31:20.328 W. Europe Standard Time"]
[GameStartTime "2018-02-14T23:10:00.128 W. Europe Standard Time"]
[PlyCount "179"]
[SetUp "1"]
[Termination "adjudication"]
[TimeControl "14400+30"]

1... Qc1+ {+1.67/51 461s} 2. Kf2 {-1.52/40 450s} g5 {+1.67/50 117s}
3. Nb5 {-1.21/37 133s} Qd1 {+1.67/52 209s} 4. Nd4 {-1.21/37 43s}
Kh7 {+1.66/51 255s} 5. Re1 {-1.27/39 353s} Qd2+ {+1.78/53 124s}
6. Re2 {-1.27/39 0.13s} Qc1 {+1.66/56 243s} 7. Rc2 {-1.25/39 309s}
Qb1 {+1.66/59 320s} 8. Rd2 {-1.23/39 488s} Kg6 {+1.65/51 196s}
9. Re2 {-1.22/41 211s} Qc1 {+1.59/61 268s} 10. Rc2 {-1.23/42 243s}
Qb1 {+1.59/62 363s} 11. Re2 {-0.01/96 153s} a6 {+2.01/54 1491s}
12. Rc2 {-1.22/44 1695s} Kf6 {+1.66/54 382s} 13. Re2 {-1.24/43 964s}
Qc1 {+1.66/52 112s} 14. Kg3 {-1.23/39 278s} Kg7 {+1.89/58 1980s}
15. Kf2 {-1.23/42 335s} b5 {+2.01/52 115s} 16. Nf5+ {-1.23/43 344s}
Kf6 {+2.01/54 91s} 17. Nd4 {-1.32/41 429s} Qa1 {+2.01/56 132s}
18. Ke3 {-1.23/40 463s} Ke5 {+2.05/50 458s} 19. f4+ {-1.26/36 381s}
gxf4+ {+2.17/44 645s} 20. Kf3+ {-1.21/37 292s} Kd6 {+2.17/44 387s}
21. Kxf4 {-1.33/36 339s} Qc1+ {+2.17/46 115s} 22. Kg4 {-1.25/34 260s}
Qf1 {+1.63/49 1416s} 23. h4 {-0.98/34 159s} Kc5 {+1.63/46 69s}
24. h5 {-0.98/36 243s} a5 {+1.63/48 137s} 25. g3 {-0.98/36 111s}
Qd1 {+2.46/36 86s} 26. Kg5 {-0.98/39 437s} Qc1+ {+2.82/39 183s}
27. Kf5 {-0.90/34 156s} b4 {+3.02/39 68s} 28. axb4+ {-0.87/32 47s}
axb4 {+3.10/46 212s} 29. Rc2 {-0.87/32 0.35s} Qh1 {+2.85/47 206s}
30. g4 {-1.58/38 496s} Qe4+ {+2.85/46 137s} 31. Kg5 {-1.58/38 0.51s}
Qe3+ {+2.85/48 172s} 32. Kh4 {-1.67/39 471s} bxc3 {+2.85/50 233s}
33. bxc3 {-1.67/39 0.18s} f6 {+2.85/53 357s} 34. Nf5 {-1.65/40 225s}
Qe4 {+2.85/49 92s} 35. Nd4 {-1.65/40 38s} f5 {+2.85/50 104s}
36. Kg5 {-1.65/39 0.13s} Qxg4+ {+2.85/52 158s} 37. Kf6 {-1.65/43 263s}
f4 {+2.77/56 311s} 38. h6 {-1.65/37 37s} Qh3 {+2.77/54 92s}
39. Kg5 {-1.58/38 175s} f3 {+2.70/55 135s} 40. h7 {-1.58/41 462s}
Qxh7 {+2.70/58 106s} 41. Rf2 {-1.58/41 0.47s} Qb1 {+2.70/62 83s}
42. Rxf3 {-1.40/38 134s} Qc1+ {+2.70/57 85s} 43. Kg6 {-1.40/41 182s}
Qe1 {+2.70/58 83s} 44. Kf7 {-1.40/43 162s} Kb6 {+2.70/63 77s}
45. Rf6+ {-1.40/37 152s} Kb7 {+2.70/67 76s} 46. Rf3 {-1.40/39 112s}
Qe5 {+2.70/68 105s} 47. Rf2 {-0.80/35 127s} Qg3 {+2.70/68 95s}
48. Rf3 {-0.80/33 0.11s} Qe1 {+2.70/69 79s} 49. Kg7 {-1.33/38 312s}
Qe5+ {+2.70/66 102s} 50. Kf7 {-1.29/38 250s} Kc8 {+2.70/64 93s}
51. Rf5 {-1.25/36 386s} Qd6 {+2.70/66 167s} 52. Rf2 {-1.26/33 168s}
Qd7+ {+2.70/65 57s} 53. Kf6 {-0.01/31 81s} Qd6+ {+2.70/67 71s}
54. Kf7 {-0.01/99 9.9s} Qg3 {+2.70/65 75s} 55. Rf3 {-1.13/30 376s}
Qe1 {+2.70/66 67s} 56. Kg7 {-1.11/39 100s} Qe5+ {+2.70/63 85s}
57. Kg6 {-1.03/30 104s} Qe4+ {+2.70/62 88s} 58. Kg7 {-0.90/24 79s}
Qe5+ {+2.70/66 100s} 59. Kg6 {-0.01/99 10s} Qe1 {+2.70/68 111s}
60. Kg7 {-0.01/94 110s} Qe7+ {+2.70/66 88s} 61. Kg6 {-0.37/30 219s}
Qe4+ {+2.70/66 132s} 62. Kg7 {-0.01/99 12s} Qe7+ {+2.70/65 154s}
63. Kg6 {-0.01/99 51s} Qd6+ {+2.70/65 116s} 64. Kf5 {-0.08/37 192s}
Kb7 {+2.70/58 60s} 65. Re3 {-0.02/34 78s} Qf8+ {+2.70/57 54s}
66. Kg4 {-0.02/32 17s} Qf6 {+2.70/57 50s} 67. Re6 {-0.01/41 81s}
Qf7 {+2.70/57 76s} 68. Re2 {-0.01/40 16s} Kc7 {+2.70/57 168s}
69. Re6 {-0.01/42 85s} Kc8 {+2.70/57 47s} 70. Re3 {-0.01/47 105s}
Kb7 {+2.70/54 44s} 71. Re2 {-0.01/74 101s} Kc7 {+2.70/42 154s}
72. Re6 {-0.01/78 197s} Kc8 {+2.70/49 33s} 73. Re3 {-0.01/81 85s}
Qf8 {+1.71/43 209s} 74. Rf3 {-0.01/43 73s} Qe7 {+2.70/47 17s}
75. Kf4 {-0.01/42 80s} Qe4+ {+2.70/56 68s} 76. Kg3 {-0.01/44 71s}
Qe1+ {+2.70/43 69s} 77. Kg4 {-0.01/48 74s} Qe7 {+2.69/38 36s}
78. Kg3 {-0.01/45 74s} Qe1+ {+2.70/33 47s} 79. Kh3 {-0.01/47 61s}
Qg1 {+2.70/47 383s} 80. Rf8+ {-0.01/48 62s} Kc7 {+2.46/42 116s}
81. Rf7+ {-0.01/44 75s} Kb6 {+2.36/37 66s} 82. Rf3 {-0.01/46 55s}
Qg6 {+1.42/40 114s} 83. Rg3 {-0.01/49 64s} Qe8 {+1.42/44 29s}
84. Kg2 {-0.01/50 67s} Qe4+ {+0.40/46 101s} 85. Kh3 {-0.01/53 57s}
Qh1+ {+1.20/40 20s} 86. Kg4 {-0.01/11 0s} Qe1 {+0.49/42 44s}
87. Rf3 {-0.01/52 69s} Qe5 {0.00/61 19s} 88. Kh3 {-0.01/51 52s}
Qh5+ {0.00/81 16s} 89. Kg2 {-0.01/66 123s} Kb7 {0.00/92 15s}
90. Re3 {-0.01/67 44s} Qg5+ {0.00/127 16s, Draw by adjudication} 1/2-1/2

[/pgn]

User avatar
Nordlandia
Posts: 2476
Joined: Fri Sep 25, 2015 7:38 pm
Location: Sortland, Norway

Re: Latest SF dev beats H6 Pro in the Goldin vs Korzubov end

Post by Nordlandia » Fri Feb 16, 2018 10:52 am

40 games à la 7+7

[d]5k2/ppp2ppp/1b6/8/6b1/2P5/PP4PP/RN2K3 w - - 0 1

CPU: i5-5200U - 1 cpu used
Hash: 512 Mb allocated per engine
EGTB: Probing 5-men - Automatic 6-men adjudication | Syzygy
GUI: CuteChess
Houdini contemt: 0
Houdini EGTB Probe Depth: 0
Stockfish: default values

Image

Game download [PGN] -> http://www103.zippyshare.com/v/qh7hzM5F/file.html

Lyudmil Tsvetkov
Posts: 6052
Joined: Tue Jun 12, 2012 10:41 am

Re: Latest SF dev beats H6 Pro in the Goldin vs Korzubov end

Post by Lyudmil Tsvetkov » Fri Feb 16, 2018 12:24 pm

Nordlandia wrote:40 games à la 7+7

[d]5k2/ppp2ppp/1b6/8/6b1/2P5/PP4PP/RN2K3 w - - 0 1

CPU: i5-5200U - 1 cpu used
Hash: 512 Mb allocated per engine
EGTB: Probing 5-men - Automatic 6-men adjudication | Syzygy
GUI: CuteChess
Houdini contemt: 0
Houdini EGTB Probe Depth: 0
Stockfish: default values

Image

Game download [PGN] -> http://www103.zippyshare.com/v/qh7hzM5F/file.html
SF is still the stronger engine, by maybe 30-50 elos, so you should be careful with that.
Yes, both positions are borderline ones, we need a perfect player playing himself to be perfectly certain.
BB + pawn vs RN would be fully equal in most cases, but here white is to play, one tempo more, plus very late endgame, where rooks are very strong.
So basically BBP vs RN is draw, this is a bit peculiar.

Lyudmil Tsvetkov
Posts: 6052
Joined: Tue Jun 12, 2012 10:41 am

Re: Latest SF dev beats H6 Pro in the Goldin vs Korzubov end

Post by Lyudmil Tsvetkov » Fri Feb 16, 2018 12:29 pm

About the Goldin-Korzubov game, you will need many LTC games, that would still be uncertain, a 10 games by a 500-elo stronger player would definitely be better.
What I observe is very strange evaluation of SF in some fortress positions, where previously the score was 200cps, now it is 300.
Obviously, this is good for game play, space/mobility?, but not for fortress detection.
But again, we don't know how this actually plays out.
What might transpire from the tests conducted is that both positions are not obvious draws.

User avatar
Nordlandia
Posts: 2476
Joined: Fri Sep 25, 2015 7:38 pm
Location: Sortland, Norway

Re: Latest SF dev beats H6 Pro in the Goldin vs Korzubov end

Post by Nordlandia » Fri Feb 16, 2018 1:22 pm

Lyudmil Tsvetkov wrote:About the Goldin-Korzubov game, you will need many LTC games, that would still be uncertain, a 10 games by a 500-elo stronger player would definitely be better.
What I observe is very strange evaluation of SF in some fortress positions, where previously the score was 200cps, now it is 300.
Obviously, this is good for game play, space/mobility?, but not for fortress detection.
But again, we don't know how this actually plays out.
What might transpire from the tests conducted is that both positions are not obvious draws.
Carlsen's intuition can give us definite answer.

I want to test BBP vs RN with black to move first . This time 20 games à la 10+10 TC.

User avatar
Nordlandia
Posts: 2476
Joined: Fri Sep 25, 2015 7:38 pm
Location: Sortland, Norway

Re: Latest SF dev beats H6 Pro in the Goldin vs Korzubov end

Post by Nordlandia » Fri Feb 16, 2018 9:56 pm

Lyudmil Tsvetkov: another borderline endgame from Kasparov Karpov 1990 match.

Looks like draw in my view. RB vs NB of opposite-color is somewhat drawish.

[d]6k1/6p1/4p1n1/3p4/2b3P1/5P1R/1B5K/8 b - - 0 4

User avatar
Nordlandia
Posts: 2476
Joined: Fri Sep 25, 2015 7:38 pm
Location: Sortland, Norway

Re: Latest SF dev beats H6 Pro in the Goldin vs Korzubov end

Post by Nordlandia » Fri Feb 16, 2018 11:22 pm

Goldin vs Korzubov match of 100 games à la 5+5 just finished. Surprisingly Houdini won more games than Stockfish!

Goldin vs Korzubov 2018


1 Houdini 6 +39 +23/=44/-14 55.56% 45.0/81
2 stockfish_x64 -39 +14/=44/-23 44.44% 36.0/81

Image

Game download [PGN] http://www1.zippyshare.com/v/PRTGXKAY/file.html


Image

Post Reply