Komodo 9.1 vs. latest SF dev.

[lkaufman]

Update: After 3404 (out of 4000) games Komodo's lead has shrunk (as I expected) to 12 elo. Score is 1759 to 1645.

[shrapnel]

I'm still not convinced that the final result of a large number of STC games can be extrapolated/compared to LTC games.
To my way of thinking, an LTC game is an LTC game and an STC game is an STC game....both test different strengths and weaknesses of Engines, and never the Twain shall meet !
Anyway, to each his own....

[lkaufman]

I'm still not convinced that the final result of a large number of STC games can be extrapolated/compared to LTC games.
To my way of thinking, an LTC game is an LTC game and an STC game is an STC game....both test different strengths and weaknesses of Engines, and never the Twain shall meet !
Anyway, to each his own....

There are certainly differences in performance of engines at different time controls. This was especially true with Houdini and the other Ippolits, which were much better at short tc. Until now, Komodo has outperformed Stockfish only at longer time controls; it is my hope that Komodo 9.1 will outperform Stockfish at all reasonable time controls, at least with four or more cores. Since Komodo has always scaled better than other engines with more time, it seems likely that if Komodo can beat SF at 2' + 1", it should also do so with more time.

[shrapnel]

Since Komodo has always scaled better than other engines with more time, it seems likely that if Komodo can beat SF at 2' + 1", it should also do so with more time.

Fair enough.
Anyway, here is another game where Stockfish either does not see or seems to under-estimate the strength of a Komodo king-side attack, until its too late.
This is the second game in which this is happening !
I wonder if the Komodo Team discovered a weakness in Stockfish's King Safety Evaluation and are exploiting it to the fullest ?
(Hope I'm not revealing any secrets ! )
[pgn][Event "120s/Move"]
[Site "www.come2play.com"]
[Date "2015.06.24"]
[Round "?"]
[White "Komodo 9.1 64-bit"]
[Black "Stockfish"]
[Result "1-0"]
[ECO "D37"]
[PlyCount "105"]

{16384MB, 1337chess PRO-220615 -plk8x74w -id4611.ctg, Anil} 1. d4 {B 0} Nf6 {77
} 2. c4 {B 0} e6 {25} 3. Nf3 {B 0} d5 {25} 4. Nc3 {B 0} Be7 {71} 5. Bf4 {B 0}
O-O {64} 6. e3 {B 0} Nbd7 {35} 7. c5 {B 0} Nh5 {39} 8. Bd3 {B 0} Nxf4 {59} 9.
exf4 {B 0} b6 {43} 10. b4 {B 0} a5 {62} 11. a3 {B 0} c6 {53} 12. O-O {B 0} Qc7
{36} 13. g3 {B 0} Ba6 {32} 14. Bxa6 {B 0} Rxa6 {6} 15. Qe2 {B 0} Raa8 {30} 16.
Na4 {B 0} b5 {125} 17. Nb2 {B 0} Bf6 {73} 18. Rfb1 {0.11/32 97} axb4 {0.12/28 6
} 19. axb4 {0.08/33 77} Ra7 {0.10/25 1} 20. Nd3 {0.09/32 83} Rfa8 {0.13/25 1}
21. Rxa7 {0.04/36 91} Qxa7 {0.05/22 0} 22. Nde5 {0.04/36 148} Qb7 23. Kg2 {0.
19/32 102} Bd8 {0.11/28 25} 24. h4 {0.36/31 105} Ra4 25. f5 {0.51/35 87} Nf8 {
0.43/30 5} 26. fxe6 {0.46/34 111} Nxe6 {0.26/27 5} 27. Re1 {0.62/33 90} Ra3 28.
Ng5 {0.66/34 89} Bxg5 {0.73/24 0} 29. hxg5 {0.65/35 63} Ra8 30. Qg4 {0.91/32 90
} Nf8 {0.90/25 2} 31. f4 {0.83/35 93} Ra2+ {0.83/25 2} 32. Kf1 {0.83/36 95} Rh2
33. Kg1 {0.83/36 78} Ra2 {0.83/22 0} 34. f5 {0.83/38 156} Qc8 35. Kf1 {1.01/31
91} Rh2 36. Qf3 {0.82/35 86} Rd2 {0.87/25 1} 37. Qf4 {0.78/37 106} Rh2 {0.78/
27 1} 38. Nd3 {0.78/34 89} Rh1+ {0.78/24 1} 39. Kf2 {0.78/36 96} Rxe1 {0.85/27
8} 40. Kxe1 {0.78/30 56} Qd8 41. Kd2 {0.78/35 102} Qe7 {0.77/33 21} 42. g4 {0.
99/33 104} Qd8 {0.98/25 1} 43. Kc3 {1.16/32 116} Qa8 {1.10/31 25} 44. Qc7 {1.
19/31 85} Qa1+ {1.07/29 8} 45. Kd2 {1.27/33 74} Qxd4 {1.20/26 3} 46. Qxc6 {1.
18/33 77} Qc4 {1.22/24 1} 47. Qe8 {1.03/34 71} Qa2+ {1.14/29 14} 48. Ke3 {1.11/
34 77} Qc4 {0.95/27 1} 49. c6 {1.11/34 77} Qxg4 50. Qxf7+ {1.50/41 84} Kxf7 {
1.57/36 7} 51. Ne5+ {2.14/37 29} Ke7 {2.16/38 24} 52. Nxg4 {1.78/39 59} Kd6 {
2.02/35 10} 53. Ne5 {2.08/39 16} (53. Ne5 {1.64/43 28} g6 {1.42/37 11} 54. f6 {
1.52/40 36} Ne6 {1.59/36 10} 55. f7 {1.65/41 45} Ke7 {1.58/36 5} 56. Kd2 {1.82/
39 98} Kf8 {1.48/37 5} 57. Ke2 {3.67/37 93} Kg7 58. Ke3 {4.02/31 22} Kf8 {4.70/
31 14} 59. Nd3 {4.17/33 21} Kxf7 60. Nf4 {6.36/31 19} d4+ {6.27/32 35} 61. Ke4
{10.43/28 60}) 1-0[/pgn]

[lkaufman]

Final result on 4 cores was 1954.5 to 1840.5, or plus eleven elo for Komodo 9.1. I then ran a 2 core match under same conditions which Komodo 9.1 won by 1005 to 995 or plus 2 elo. So on one or two cores they are very close, but by 4 cores Komodo pulls ahead by a significant margin.

[Dr.Wael Deeb]

Since Komodo has always scaled better than other engines with more time, it seems likely that if Komodo can beat SF at 2' + 1", it should also do so with more time.

Fair enough.
Anyway, here is another game where Stockfish either does not see or seems to under-estimate the strength of a Komodo king-side attack, until its too late.
This is the second game in which this is happening !
I wonder if the Komodo Team discovered a weakness in Stockfish's King Safety Evaluation and are exploiting it to the fullest ?
(Hope I'm not revealing any secrets ! )
[pgn][Event "120s/Move"]
[Site "www.come2play.com"]
[Date "2015.06.24"]
[Round "?"]
[White "Komodo 9.1 64-bit"]
[Black "Stockfish"]
[Result "1-0"]
[ECO "D37"]
[PlyCount "105"]

{16384MB, 1337chess PRO-220615 -plk8x74w -id4611.ctg, Anil} 1. d4 {B 0} Nf6 {77
} 2. c4 {B 0} e6 {25} 3. Nf3 {B 0} d5 {25} 4. Nc3 {B 0} Be7 {71} 5. Bf4 {B 0}
O-O {64} 6. e3 {B 0} Nbd7 {35} 7. c5 {B 0} Nh5 {39} 8. Bd3 {B 0} Nxf4 {59} 9.
exf4 {B 0} b6 {43} 10. b4 {B 0} a5 {62} 11. a3 {B 0} c6 {53} 12. O-O {B 0} Qc7
{36} 13. g3 {B 0} Ba6 {32} 14. Bxa6 {B 0} Rxa6 {6} 15. Qe2 {B 0} Raa8 {30} 16.
Na4 {B 0} b5 {125} 17. Nb2 {B 0} Bf6 {73} 18. Rfb1 {0.11/32 97} axb4 {0.12/28 6
} 19. axb4 {0.08/33 77} Ra7 {0.10/25 1} 20. Nd3 {0.09/32 83} Rfa8 {0.13/25 1}
21. Rxa7 {0.04/36 91} Qxa7 {0.05/22 0} 22. Nde5 {0.04/36 148} Qb7 23. Kg2 {0.
19/32 102} Bd8 {0.11/28 25} 24. h4 {0.36/31 105} Ra4 25. f5 {0.51/35 87} Nf8 {
0.43/30 5} 26. fxe6 {0.46/34 111} Nxe6 {0.26/27 5} 27. Re1 {0.62/33 90} Ra3 28.
Ng5 {0.66/34 89} Bxg5 {0.73/24 0} 29. hxg5 {0.65/35 63} Ra8 30. Qg4 {0.91/32 90
} Nf8 {0.90/25 2} 31. f4 {0.83/35 93} Ra2+ {0.83/25 2} 32. Kf1 {0.83/36 95} Rh2
33. Kg1 {0.83/36 78} Ra2 {0.83/22 0} 34. f5 {0.83/38 156} Qc8 35. Kf1 {1.01/31
91} Rh2 36. Qf3 {0.82/35 86} Rd2 {0.87/25 1} 37. Qf4 {0.78/37 106} Rh2 {0.78/
27 1} 38. Nd3 {0.78/34 89} Rh1+ {0.78/24 1} 39. Kf2 {0.78/36 96} Rxe1 {0.85/27
8} 40. Kxe1 {0.78/30 56} Qd8 41. Kd2 {0.78/35 102} Qe7 {0.77/33 21} 42. g4 {0.
99/33 104} Qd8 {0.98/25 1} 43. Kc3 {1.16/32 116} Qa8 {1.10/31 25} 44. Qc7 {1.
19/31 85} Qa1+ {1.07/29 8} 45. Kd2 {1.27/33 74} Qxd4 {1.20/26 3} 46. Qxc6 {1.
18/33 77} Qc4 {1.22/24 1} 47. Qe8 {1.03/34 71} Qa2+ {1.14/29 14} 48. Ke3 {1.11/
34 77} Qc4 {0.95/27 1} 49. c6 {1.11/34 77} Qxg4 50. Qxf7+ {1.50/41 84} Kxf7 {
1.57/36 7} 51. Ne5+ {2.14/37 29} Ke7 {2.16/38 24} 52. Nxg4 {1.78/39 59} Kd6 {
2.02/35 10} 53. Ne5 {2.08/39 16} (53. Ne5 {1.64/43 28} g6 {1.42/37 11} 54. f6 {
1.52/40 36} Ne6 {1.59/36 10} 55. f7 {1.65/41 45} Ke7 {1.58/36 5} 56. Kd2 {1.82/
39 98} Kf8 {1.48/37 5} 57. Ke2 {3.67/37 93} Kg7 58. Ke3 {4.02/31 22} Kf8 {4.70/
31 14} 59. Nd3 {4.17/33 21} Kxf7 60. Nf4 {6.36/31 19} d4+ {6.27/32 35} 61. Ke4
{10.43/28 60}) 1-0[/pgn]

Not so many moons ago it was just the other way around....

An amazing achievement of the Komodo team....
Dr.D

[ouachita]

Anil, I still haven't gotten over your post:

Stockfish is for little boys playing quick games on their laptops or crummy AMD hardware!

And here I thought my sarcasm was unequalled.

[S.Taylor]

I'm still not convinced that the final result of a large number of STC games can be extrapolated/compared to LTC games.
To my way of thinking, an LTC game is an LTC game and an STC game is an STC game....both test different strengths and weaknesses of Engines, and never the Twain shall meet !
Anyway, to each his own....

There are certainly differences in performance of engines at different time controls. This was especially true with Houdini and the other Ippolits, which were much better at short tc. Until now, Komodo has outperformed Stockfish only at longer time controls; it is my hope that Komodo 9.1 will outperform Stockfish at all reasonable time controls, at least with four or more cores. Since Komodo has always scaled better than other engines with more time, it seems likely that if Komodo can beat SF at 2' + 1", it should also do so with more time.

Even it is weaker on low speed cpu/cores than SF, but stronger, the longer/more cores it has, then i don't personally care one bit about it losing as few elo in blitz.

It will always be much stronger than any human in blitz or fast TC's anyway. And since my main interest in computer is to see what move it makes after 10 minutes/30 minutes or even a lot longer, then Komodo would be clearly THE most reliable engine to find the best move, if that's what you are looking for.

(But i have not recently found out how to put todays engines on infinite and move pieces around and see it's top lines and updatings instantly. But i take it that's only MY problem).