Testing Komodo 6 : first 110 games

[beram]

I ve tested my first 110 games of Komodo 6 MP against Houdini 3
Komodo 6 scores 46,36 %
That is 7,7 %(!) better than previous tests with Komodo 5.1 MP

Next test is running now against Stockfish 0510

Code: Select all

AMD X6 1090T 3200 both, 4CPU, ponder off, TC 3m+2s
Bram testsuite 3.0 - 50 games
                   
1   Houdini 3 x64    +17/=18/-15 52.00%   26.0/50
2   Komodo 6 64-bit  +15/=18/-17 48.00%   24.0/50

60 games Noomen Testsuite 2012
                   
1   Houdini 3 x64    +18/=30/-12 55.00%   33.0/60
2   Komodo 6 64-bit  +12/=30/-18 45.00%   27.0/60

110 games  overall

Houdini 3- Komodo 6, Blitz 3m+2s  
                     
1   Houdini 3 x64    +35/=48/-27 53.64%   59.0/110
2   Komodo 6 64-bit  +27/=48/-35 46.36%   51.0/110

Code: Select all

Houdini3 - Komodo5.1, 4CPU Blitz 4m+2s,  1024 Mb Hash
                       
1   Houdini 3 x64      +81/=108/-31 61.36%  135.0/220
2   Komodo 5.1 64-bit  +31/=108/-81 38.64%   85.0/220

[beram]

I ve tested my first 110 games of Komodo 6 MP against Houdini 3
Komodo 6 scores 46,36 %
That is 7,7 %(!) better than previous tests with Komodo 5.1 MP

Next test is running now against Stockfish 0510

Code: Select all

AMD X6 1090T 3200 both, 4CPU, ponder off, TC 3m+2s
Bram testsuite 3.0 - 50 games
                   
1   Houdini 3 x64    +17/=18/-15 52.00%   26.0/50
2   Komodo 6 64-bit  +15/=18/-17 48.00%   24.0/50

60 games Noomen Testsuite 2012
                   
1   Houdini 3 x64    +18/=30/-12 55.00%   33.0/60
2   Komodo 6 64-bit  +12/=30/-18 45.00%   27.0/60

110 games  overall

Houdini 3- Komodo 6, Blitz 3m+2s  
                     
1   Houdini 3 x64    +35/=48/-27 53.64%   59.0/110
2   Komodo 6 64-bit  +27/=48/-35 46.36%   51.0/110

After next 110 games, this time against Stockfish 0810
Komodo 6 gets 45.45 %

with TC 3m 2s on my AMD X6 1090T 3200, 4CPU, the 'new order' is:
1 - Stockfish 4 dev
2 - Houdini 3
3 - Komodo 6

Code: Select all

AMD X6 1090T 3200 both, 4CPU, ponder off, TC 3m+2s
Bram testsuite 3.0 - 50 games
                              
1   Komodo 6 64-bit             +13/=26/-11 52.00%   26.0/50
2   Stockfish 081013 64 SSE4.2  +11/=26/-13 48.00%   24.0/50

60 games Noomen Testsuite 2012
                               
1   Stockfish 081013 64 SSE4.2  +20/=32/-8 60.00%   36.0/60
2   Komodo 6 64-bit             +8/=32/-20 40.00%   24.0/60

110 games  overall

Komodo 6 -Stockfish 0810, Blitz 3m+2s 
                               
1   Stockfish 081013 64 SSE4.2  +31/=58/-21 54.55%   60.0/110
2   Komodo 6 64-bit             +21/=58/-31 45.45%   50.0/110

[Dr.Wael Deeb]

Thanks Bram for the results

I am almost convinced now that there is no clear leader at the top of the computer chess mountain....

Certin chess engines are shifting their positions accordingly and unless a new much improved version of Houdini is released,the situation remains more or less the same....

Dr.D

[beram]

Yet another match between new Komodo 6 and the latest SF dev 0910

Code: Select all

Komodo 6 -Stockfish 0910, 4CPU, Blitz 5m+3s
AMD X6 1090T 3200 Mhz, (Fritz, 2013.10.10 - 2013.10.11
                                  Score     12345678901234567890123456789012345678901234567890
-----------------------------------------------------------------------------------------------
 1: Stockfish 091013 64 SSE4.2  29.5 / 50   ==0==011=1=10110=101010111==101101=0=110==10=101=1  (+22 -13 =15)
 2: Komodo 6 64-bit             20.5 / 50   ==1==100=0=01001=010101000==010010=1=001==01=010=0  (+13 -22 =15)
-----------------------------------------------------------------------------------------------
50 games: +22 =15 -13

Two game examples for both programs
First is a typical win for Komodo where it shows its superiority in games
with unbalanced material when it has the pieces against the queen
SF and Houdini both dont see the black advantage for the queen sac at move 17 for many moves


Second one is a crazy and beautiful game where the white king marches to g8! and the black king to b1!
First SF sacs a piece for two pawns and compensation, than nice positional play
followed by a king march to the field g8! (move 53) and Komodo's king marches up to the opposite field b2 (move 72)

[pgn]
[Event "Komodo 6 -Stockfish 0510, Blitz 5m+3s"]
[Site "AMD X6 1090T 3200 Mhz, (Fritz"]
[Date "2013.10.11"]
[Round "30"]
[White "Stockfish 091013 64 SSE4.2"]
[Black "Komodo 6 64-bit"]
[Result "0-1"]
[ECO "B90"]
[Annotator "0.00;0.30"]
[PlyCount "138"]
[TimeControl "300+3"]

{AMD Phenom(tm) II X6 1090T Processor 3207 MHz W=25.5 plies; 5.774kN/s B=20.
8 plies; 4.574kN/s} 1. e4 c5 2. Nf3 d6 3. d4 cxd4 4. Nxd4 a6 5. Nc3 Nf6 6. Be3
Ng4 7. Bg5 h6 8. Bh4 g5 9. Bg3 Bg7 10. Be2 h5 {Both last book move} 11. Bxg4 {
[%eval 0,24] [%emt 0:00:11]} Bxg4 {[%eval 30,19] [%emt 0:00:12]} 12. f3 {
[%eval 0,25] [%emt 0:00:07]} Bd7 {[%eval 28,20] [%emt 0:00:15]} 13. Bf2 {
[%eval 2,24] [%emt 0:00:06]} Nc6 {[%eval 29,20] [%emt 0:00:10]} 14. O-O {
[%eval 10,25] [%emt 0:00:06]} Rc8 {[%eval 30,20] [%emt 0:00:25] (h4)} 15. Nd5 {
[%eval 22,24] [%emt 0:00:08]} e6 {[%eval 31,20] [%emt 0:00:09] (0-0)} 16. Nf5 {
[%eval 66,26] [%emt 0:00:07] (Nxc6)} exf5 {[%eval -16,20] [%emt 0:00:09]} 17.
Bb6 {[%eval 48,26] [%emt 0:00:07]} Qxb6+ {[%eval -23,20] [%emt 0:00:06]} 18.
Nxb6 {[%eval 68,26] [%emt 0:00:03]} Bd4+ {[%eval 2,19] [%emt 0:00:06]} 19. Kh1
{[%eval 58,27] [%emt 0:00:13]} Bxb6 {[%eval -19,19] [%emt 0:00:04]} 20. Qxd6 {
[%eval 60,28] [%emt 0:00:07]} fxe4 {[%eval -17,21] [%emt 0:00:12]} 21. Qf6 {
[%eval 48,29] [%emt 0:00:55]} Rg8 {[%eval -13,21] [%emt 0:00:17]} 22. fxe4 {
[%eval 42,28] [%emt 0:00:11]} Be6 {[%eval -14,21] [%emt 0:00:04]} 23. Rad1 {
[%eval 54,28] [%emt 0:00:27]} Bc7 {[%eval -34,21] [%emt 0:00:12]} 24. Qf2 {
[%eval 34,27] [%emt 0:00:45] (Rd5)} Be5 {[%eval -52,17] [%emt 0:00:06] (h4)}
25. Qb6 {[%eval 26,23] [%emt 0:00:07]} Rc7 {[%eval -25,18] [%emt 0:00:03] (Rb8)
} 26. b4 {[%eval 82,23] [%emt 0:00:09]} Rg6 {[%eval -54,18] [%emt 0:00:05] (g4)
} 27. a4 {[%eval 36,23] [%emt 0:00:06]} Bd7 {[%eval -31,19] [%emt 0:00:16]
(Ne7)} 28. Qe3 {[%eval 32,24] [%emt 0:00:27] (Rf2)} Ne7 {[%eval -65,18] [%emt
0:00:05]} 29. Qd3 {[%eval 0,25] [%emt 0:00:15]} f6 {[%eval -79,21] [%emt 0:00:
16] (h4)} 30. b5 {[%eval 0,23] [%emt 0:00:05] (Qb3)} axb5 {[%eval -79,21]
[%emt 0:00:08]} 31. axb5 {[%eval 0,26] [%emt 0:00:07]} Bg4 {[%eval -85,20]
[%emt 0:00:06]} 32. Rc1 {[%eval 0,28] [%emt 0:00:20]} Rc3 {[%eval -91,19]
[%emt 0:00:05]} 33. Qd2 {[%eval 0,28] [%emt 0:00:07]} Kf7 {[%eval -95,18]
[%emt 0:00:04] (h4)} 34. Ra1 {[%eval -18,21] [%emt 0:00:07]} Rg8 {[%eval -101,
19] [%emt 0:00:05]} 35. Ra5 {[%eval -42,23] [%emt 0:00:08] (Ra7)} Rc5 {[%eval
-109,19] [%emt 0:00:07] (Rcc8)} 36. Ra7 {[%eval -54,23] [%emt 0:00:07] (Qe3)}
Rc7 {[%eval -115,19] [%emt 0:00:10]} 37. Qe3 {[%eval -80,22] [%emt 0:00:07]}
Rd7 {[%eval -124,17] [%emt 0:00:03] (Be6)} 38. Ra3 {[%eval -38,22] [%emt 0:00:
05]} Nc8 {[%eval -124,18] [%emt 0:00:04] (Rc8)} 39. h3 {[%eval -119,23] [%emt
0:00:31] (Rd3)} Be6 {[%eval -158,18] [%emt 0:00:09]} 40. Rc3 {[%eval -155,23]
[%emt 0:00:16] (Rd3)} Nd6 {[%eval -168,20] [%emt 0:00:12]} 41. Rc5 {[%eval
-204,22] [%emt 0:00:05]} Kg6 {[%eval -176,17] [%emt 0:00:02]} 42. Rxe5 {[%eval
-210,25] [%emt 0:00:16] (Qf3)} fxe5 {[%eval -177,20] [%emt 0:00:03]} 43. Qc5 {
[%eval -224,25] [%emt 0:00:04]} Nc4 {[%eval -177,21] [%emt 0:00:04]} 44. Rf8 {
[%eval -230,25] [%emt 0:00:04]} Rxf8 {[%eval -210,21] [%emt 0:00:07]} 45. Qxf8
{[%eval -276,24] [%emt 0:00:03]} Nd6 {[%eval -211,20] [%emt 0:00:03]} 46. Qh8 {
[%eval -284,24] [%emt 0:00:03]} Nf7 {[%eval -210,20] [%emt 0:00:03]} 47. Qa8 {
[%eval -312,24] [%emt 0:00:04]} g4 {[%eval -225,19] [%emt 0:00:04]} 48. Kh2 {
[%eval -312,24] [%emt 0:00:03]} gxh3 {[%eval -225,19] [%emt 0:00:03]} 49. gxh3
{[%eval -307,25] [%emt 0:00:02]} Rd2+ {[%eval -227,21] [%emt 0:00:06]} 50. Kg1
{[%eval -307,25] [%emt 0:00:02]} Rxc2 {[%eval -225,22] [%emt 0:00:09]} 51. h4 {
[%eval -329,27] [%emt 0:00:04]} Rc1+ {[%eval -244,24] [%emt 0:00:20]} 52. Kg2 {
[%eval -337,27] [%emt 0:00:05]} Rc7 {[%eval -244,24] [%emt 0:00:04] (Bc8)} 53.
Qa7 {[%eval -335,23] [%emt 0:00:02] (Kg1)} Kg7 {[%eval -244,25] [%emt 0:00:05]
(Rc2+)} 54. Qb6 {[%eval -337,26] [%emt 0:00:02]} Re7 {[%eval -244,24] [%emt 0:
00:04]} 55. Qg1 {[%eval -353,27] [%emt 0:00:02] (Qc5)} Nh8 {[%eval -259,25]
[%emt 0:00:06] (Bg4)} 56. Qd1 {[%eval -343,26] [%emt 0:00:03] (Qc5)} Ng6 {
[%eval -259,24] [%emt 0:00:07] (Bg4)} 57. Kg3 {[%eval -343,29] [%emt 0:00:03]}
Bg4 {[%eval -261,24] [%emt 0:00:04]} 58. Qc1 {[%eval -343,28] [%emt 0:00:03]
(Qd5)} Be2 {[%eval -324,22] [%emt 0:00:10] (Rd7)} 59. Qc5 {[%eval -426,27]
[%emt 0:00:02]} Rd7 {[%eval -324,23] [%emt 0:00:04] (Rf7)} 60. Kf2 {[%eval
-507,25] [%emt 0:00:04] (b6)} Bd3 {[%eval -370,23] [%emt 0:00:05]} 61. Ke3 {
[%eval -535,28] [%emt 0:00:06]} Kh7 {[%eval -377,22] [%emt 0:00:05]} 62. b6 {
[%eval -543,28] [%emt 0:00:03] (Qb6)} Kg7 {[%eval -411,25] [%emt 0:00:03]} 63.
Qc1 {[%eval -551,30] [%emt 0:00:02]} Ba6 {[%eval -445,24] [%emt 0:00:07] (Bb5)}
64. Qc5 {[%eval -545,26] [%emt 0:00:05] (Qg1)} Rf7 {[%eval -474,22] [%emt 0:00:
02] (Bf1)} 65. Qd6 {[%eval -545,25] [%emt 0:00:02] (Qd5)} Rf4 {[%eval -474,25]
[%emt 0:00:04]} 66. Qd5 {[%eval -650,26] [%emt 0:00:03]} Rxh4 {[%eval -492,21]
[%emt 0:00:03]} 67. Kf2 {[%eval -696,28] [%emt 0:00:03] (Qd7+)} Rg4 {[%eval
-492,26] [%emt 0:00:06]} 68. Qd1 {[%eval -725,29] [%emt 0:00:04] (Qd7+)} Bb5 {
[%eval -551,23] [%emt 0:00:07] (Rf4+)} 69. Ke3 {[%eval -838,26] [%emt 0:00:02]}
Bc6 {[%eval -605,22] [%emt 0:00:03]} 0-1
[/pgn]


[pgn]
[Event "Komodo 6 -Stockfish 0910, Blitz 5m+3s"]
[Site "AMD X6 1090T 3200 Mhz, (Fritz"]
[Date "2013.10.11"]
[Round "50"]
[White "Stockfish 091013 64 SSE4.2"]
[Black "Komodo 6 64-bit"]
[Result "1-0"]
[ECO "D15"]
[Annotator "0.48;0.28"]
[PlyCount "177"]
[TimeControl "300+3"]

{AMD Phenom(tm) II X6 1090T Processor 3207 MHz W=25.8 plies; 6.224kN/s B=20.
2 plies; 4.634kN/s} 1. d4 d5 2. Nf3 Nf6 3. c4 c6 4. Nc3 a6 5. Ne5 Nbd7 {Both
last book move} 6. Bf4 {[%eval 48,22] [%emt 0:00:06]} dxc4 {[%eval 28,19]
[%emt 0:00:17] (Nh5)} 7. Nxc4 {[%eval 42,23] [%emt 0:00:09]} b5 {[%eval 14,21]
[%emt 0:00:29]} 8. Ne5 {[%eval 12,24] [%emt 0:00:08]} Bb7 {[%eval 9,19] [%emt
0:00:05]} 9. e3 {[%eval 16,23] [%emt 0:00:06] (Qc2)} c5 {[%eval 5,18] [%emt 0:
00:04]} 10. dxc5 {[%eval 0,24] [%emt 0:00:18] (Nxd7)} Nxe5 {[%eval -6,18]
[%emt 0:00:06]} 11. Qxd8+ {[%eval 22,26] [%emt 0:00:13] (Bxe5)} Rxd8 {[%eval
-17,20] [%emt 0:00:03]} 12. Bxe5 {[%eval 22,26] [%emt 0:00:08]} Nd7 {[%eval 7,
22] [%emt 0:00:12]} 13. Bg3 {[%eval 20,27] [%emt 0:00:07]} Nxc5 {[%eval 27,22]
[%emt 0:00:11] (h5)} 14. Nxb5 {[%eval 60,27] [%emt 0:00:05]} axb5 {[%eval 28,
21] [%emt 0:00:06]} 15. Bxb5+ {[%eval 54,27] [%emt 0:00:08]} Nd7 {[%eval 25,21]
[%emt 0:00:04]} 16. O-O {[%eval 56,29] [%emt 0:00:10]} e5 {[%eval 31,24] [%emt
0:00:15]} 17. Rfd1 {[%eval 64,29] [%emt 0:00:06]} Ke7 {[%eval 27,22] [%emt 0:
00:06]} 18. Rac1 {[%eval 46,29] [%emt 0:00:10]} Nc5 {[%eval 25,22] [%emt 0:00:
10]} 19. Rxd8 {[%eval 50,28] [%emt 0:00:09]} Kxd8 {[%eval 29,21] [%emt 0:00:04]
} 20. Bxe5 {[%eval 50,29] [%emt 0:00:08]} f6 {[%eval 25,22] [%emt 0:00:05]
(Bc8)} 21. Bd4 {[%eval 86,27] [%emt 0:00:07]} Ba6 {[%eval 27,20] [%emt 0:00:04]
} 22. Bxa6 {[%eval 62,28] [%emt 0:00:06]} Nxa6 {[%eval 30,21] [%emt 0:00:04]}
23. Rc6 {[%eval 66,29] [%emt 0:00:08]} Nb4 {[%eval 28,22] [%emt 0:00:05]} 24.
Rxf6 {[%eval 70,29] [%emt 0:00:13]} Ke8 {[%eval 36,22] [%emt 0:00:07] (Nxa2)}
25. Rb6 {[%eval 76,26] [%emt 0:00:07]} Nxa2 {[%eval 46,21] [%emt 0:00:07]} 26.
f4 {[%eval 84,27] [%emt 0:00:06]} Rg8 {[%eval 37,21] [%emt 0:00:09]} 27. Kf2 {
[%eval 64,26] [%emt 0:00:09] (Rb8+)} Nb4 {[%eval 47,21] [%emt 0:00:12]} 28.
Rb8+ {[%eval 72,26] [%emt 0:00:10] (Ke2)} Kf7 {[%eval 43,20] [%emt 0:00:05]}
29. Rb7+ {[%eval 87,27] [%emt 0:00:06]} Ke8 {[%eval 43,22] [%emt 0:00:09] (Ke6)
} 30. Kf3 {[%eval 115,26] [%emt 0:00:06]} Nc6 {[%eval 46,23] [%emt 0:00:09]}
31. Bc3 {[%eval 113,28] [%emt 0:00:06]} Kd8 {[%eval 56,23] [%emt 0:00:07]} 32.
g4 {[%eval 119,28] [%emt 0:00:10]} Be7 {[%eval 49,23] [%emt 0:00:09]} 33. e4 {
[%eval 109,29] [%emt 0:00:22]} Kc8 {[%eval 49,22] [%emt 0:00:05] (h6)} 34. Rb5
{[%eval 117,30] [%emt 0:00:09]} Kc7 {[%eval 54,22] [%emt 0:00:15]} 35. Rf5 {
[%eval 94,26] [%emt 0:00:12] (Rh5)} Nd8 {[%eval 52,19] [%emt 0:00:04]} 36. Ba5+
{[%eval 94,25] [%emt 0:00:05] (Rh5)} Kb7 {[%eval 43,20] [%emt 0:00:14] (Kc8)}
37. Rb5+ {[%eval 103,24] [%emt 0:00:06]} Kc8 {[%eval 54,19] [%emt 0:00:03]
(Ka6)} 38. e5 {[%eval 105,27] [%emt 0:00:10] (Bc3)} Nc6 {[%eval 56,20] [%emt 0:
00:16]} 39. Bc3 {[%eval 90,28] [%emt 0:00:05]} Rd8 {[%eval 56,19] [%emt 0:00:
02] (Kc7)} 40. Ke4 {[%eval 144,25] [%emt 0:00:05]} Kc7 {[%eval 54,19] [%emt 0:
00:04] (Rd1)} 41. f5 {[%eval 159,25] [%emt 0:00:07]} Ra8 {[%eval 54,19] [%emt
0:00:07]} 42. Kd5 {[%eval 145,25] [%emt 0:00:10]} Rf8 {[%eval 62,20] [%emt 0:
00:07] (Ra4)} 43. Ke6 {[%eval 195,24] [%emt 0:00:07] (h3)} Bg5 {[%eval 71,19]
[%emt 0:00:05]} 44. Rc5 {[%eval 195,25] [%emt 0:00:05]} Kb6 {[%eval 90,18]
[%emt 0:00:08]} 45. Rd5 {[%eval 214,26] [%emt 0:00:05]} Nd8+ {[%eval 76,18]
[%emt 0:00:05]} 46. Kd7 {[%eval 230,27] [%emt 0:00:08]} Nc6 {[%eval 108,20]
[%emt 0:00:11]} 47. Rd6 {[%eval 238,28] [%emt 0:00:07] (h3)} Rf7+ {[%eval 78,
19] [%emt 0:00:04]} 48. Ke8 {[%eval 238,27] [%emt 0:00:06] (Ke6)} Re7+ {[%eval
89,20] [%emt 0:00:05]} 49. Kf8 {[%eval 238,12] [%emt 0:00:00]} Rb7 {[%eval 87,
21] [%emt 0:00:04] (Rc7)} 50. f6 {[%eval 250,25] [%emt 0:00:05]} gxf6 {[%eval
97,21] [%emt 0:00:07]} 51. exf6 {[%eval 226,25] [%emt 0:00:03]} Kc5 {[%eval
112,22] [%emt 0:00:19]} 52. Rd1 {[%eval 254,26] [%emt 0:00:07]} h6 {[%eval 115,
21] [%emt 0:00:09]} 53. Kg8 {[%eval 264,27] [%emt 0:00:13]} Rb8+ {[%eval 97,19]
[%emt 0:00:02]} 54. Kf7 {[%eval 272,27] [%emt 0:00:06] (Kg7)} Nd8+ {[%eval 118,
20] [%emt 0:00:08]} 55. Kg6 {[%eval 264,29] [%emt 0:00:07]} Ne6 {[%eval 103,18]
[%emt 0:00:01]} 56. Be5 {[%eval 272,30] [%emt 0:00:06] (f7)} Ra8 {[%eval 121,
19] [%emt 0:00:08] (Rb3)} 57. f7 {[%eval 250,24] [%emt 0:00:06]} Ra6 {[%eval
114,18] [%emt 0:00:04]} 58. Kf5 {[%eval 314,25] [%emt 0:00:05] (Rf1)} Nf8 {
[%eval 157,20] [%emt 0:00:08]} 59. Bg7 {[%eval 355,26] [%emt 0:00:05]} Kc4 {
[%eval 156,20] [%emt 0:00:03] (Be7)} 60. Rd4+ {[%eval 406,24] [%emt 0:00:05]}
Kb3 {[%eval 156,21] [%emt 0:00:03]} 61. Rd3+ {[%eval 375,24] [%emt 0:00:08]
(Rb4+)} Kc2 {[%eval 149,19] [%emt 0:00:02]} 62. Rc3+ {[%eval 395,25] [%emt 0:
00:05]} Kb1 {[%eval 152,19] [%emt 0:00:03] (Kd2)} 63. Rc7 {[%eval 383,23]
[%emt 0:00:06] (Rc8)} Ra5+ {[%eval 137,18] [%emt 0:00:02]} 64. Be5 {[%eval 396,
25] [%emt 0:00:04] (Ke4)} Rb5 {[%eval 123,21] [%emt 0:00:07]} 65. Rc8 {[%eval
369,26] [%emt 0:00:04]} Be7 {[%eval 150,20] [%emt 0:00:04]} 66. Rb8 {[%eval
406,27] [%emt 0:00:08]} Rd5 {[%eval 153,20] [%emt 0:00:04]} 67. Re8 {[%eval
427,27] [%emt 0:00:08] (Ke4)} Bb4 {[%eval 154,21] [%emt 0:00:06]} 68. h4 {
[%eval 354,28] [%emt 0:00:21] (Ke4)} Kc2 {[%eval 154,20] [%emt 0:00:13] (Rd1)}
69. Ke4 {[%eval 420,26] [%emt 0:00:04]} Rd7 {[%eval 169,21] [%emt 0:00:06]} 70.
Rb8 {[%eval 408,26] [%emt 0:00:03]} Be7 {[%eval 178,20] [%emt 0:00:04]} 71. g5
{[%eval 440,25] [%emt 0:00:03] (Bg7)} hxg5 {[%eval 123,16] [%emt 0:00:02]} 72.
hxg5 {[%eval 404,25] [%emt 0:00:03]} Kb1 {[%eval 121,19] [%emt 0:00:04]} 73.
Rc8 {[%eval 387,26] [%emt 0:00:09] (Kf5)} Bb4 {[%eval 139,19] [%emt 0:00:05]
(Ka2)} 74. Rc7 {[%eval 428,22] [%emt 0:00:03]} Rd1 {[%eval 194,19] [%emt 0:00:
06]} 75. Bc3 {[%eval 501,24] [%emt 0:00:03]} Ne6 {[%eval 241,18] [%emt 0:00:13]
(Bxc3)} 76. g6 {[%eval 583,20] [%emt 0:00:01]} Bf8 {[%eval 251,20] [%emt 0:00:
04]} 77. Rc8 {[%eval 616,25] [%emt 0:00:03] (Rc6)} Ng5+ {[%eval 260,19] [%emt
0:00:02] (Bh6)} 78. Ke5 {[%eval 616,28] [%emt 0:00:03] (Kf5)} Bg7+ {[%eval 259,
22] [%emt 0:00:04]} 79. Kf5 {[%eval 624,29] [%emt 0:00:03]} Bh6 {[%eval 258,22]
[%emt 0:00:03]} 80. Bf6 {[%eval 642,25] [%emt 0:00:04]} Rd5+ {[%eval 260,22]
[%emt 0:00:04]} 81. Be5 {[%eval 694,25] [%emt 0:00:03] (Kg4)} Nf3 {[%eval 342,
19] [%emt 0:00:08] (Rd1)} 82. Ke4 {[%eval 735,21] [%emt 0:00:01]} Rxe5+ {
[%eval 362,21] [%emt 0:00:04]} 83. Kxf3 {[%eval 850,24] [%emt 0:00:02]} Rf5+ {
[%eval 414,19] [%emt 0:00:04] (Re3+)} 84. Ke4 {[%eval 924,23] [%emt 0:00:02]}
Rf6 {[%eval 458,19] [%emt 0:00:04] (Rf1)} 85. Rh8 {[%eval 1896,23] [%emt 0:00:
03]} Bg7 {[%eval 539,20] [%emt 0:00:07]} 86. Rh7 {[%eval 2032,23] [%emt 0:00:
01] (Rg8)} Bf8 {[%eval 778,18] [%emt 0:00:10] (Rxg6)} 87. g7 {[%eval 7709,22]
[%emt 0:00:03]} Bxg7 {[%eval 779,21] [%emt 0:00:03]} 88. Rxg7 {[%eval 8666,29]
[%emt 0:00:02]} Kxb2 {[%eval 791,21] [%emt 0:00:03]} 89. Ke5 {[%eval 9402,30]
[%emt 0:00:01]} 1-0
[/pgn]

[Lyudmil Tsvetkov]

Hi Bram, thanks for the post.
Great games, even at 5+3.

How long, according to you, will it take with other eval terms and search routines apart from introducing proper imbalances until Stock is able to avoid falling into similar traps? My uneducated, layman's guess is that it will take longer than 10 years The simple truth is that no other eval terms, however well tuned, or search routines, will be able to properly solve imbalances.

On the second game: for me, Stockfish is the best endgame player of all. It is real joy to watch Stockfish play endgames. The reasons?
- Stockfish as no other engine activates its king in the endgame, very subtle, very good-looking
- as it is obvious from the Stockfish parameters, the engine scores heavily mobility in the endgame, and with a lower number of pieces it would be more important that they are all mobile than with a bigger number of pieces as in the middlegame.

2 very simple rules, but they bear fruit. Probably there are also many other search and eval Stockfish endgame subtleties, but I do not recognize them.

It is about time to start gathering info about 'The computer chess game of the century' (or at least the decade). The above 2 are not very bad first tries.

[Lyudmil Tsvetkov]

And the really nasty thing is that relatively good imbalances could be done in 5 minutes.

The formula is very simple, even Capablanca once said that queen and knight perform better than queen and bishop because they attack different squares:

- you always and separately assign the respective bonus for the pair of bishops
- independently of that, you do a very simple complementarity check
- you score bonus points for each piece type on the board, some 20-30-50cps, should be tested; the 3 piece types are linear (queen and rooks), diagonal (queen and bishops) and a knight type.
- you score a penalty of maybe some 10-15cps for each repetition of a piece type, i.e., when you have 2 knights, you penalise by 10cps, when you have queen and 2 rooks, you penalise by 20cps for 3 linear pieces and 2 repetitions, queen and bishop are penalised for one repetition of a diagonal type with 10cps, etc.
- you check the piece types and repetitions and come up with a complementarity score for the interaction of all pieces, otherwise imbalances; you always add to that the pair of bishops score

Is not the math as simple as hell? But how many engines successfully implement imbalances?

Of course, even if someone tries to do this, it will not work, very obviously why: you will have also to tune at least piece values, but that might be a hell of a job, because it will have implications on all other terms related to the current piece values. And the job could be altogether impossible, unless you know with great precision what exactly every single eval term does to change accordingly. If you however know what every eval term does, you might be able to do the tuning in another 5 minutes

So basically, it is a 10 minutes affair, or altogether impossible.

[M ANSARI]

In the first game I think that you will find that the value of the queen is probably set too high due to the queen doing well in testing at super fast time controls. Most evaluation tuning today is done by playing millions of lighting games, and the dynamic abilities of a queen do really well in those situations. However if there is enough time for the pieces to coordinate the pieces without a fatal blow being inflicted by the queen, the pieces tend to quickly overwhelm the queen. I remember early Houdini versions especially had overly high queen values, but it seemed to be toned down a little now with H3.

[Lyudmil Tsvetkov]

It is not only the queen value.
The queen, the bishop and the knight all have their respective values, but it would be useful to have, apart from them, a bonus for the way the pieces complement in a group.
For example, a small bonus for a queen and a knight versus a queen and a bishop separately from their piece values, because they complement each other better, perform better when taken together, as in the queen and knight you have 3 piece types: linear, diagonal and a knight type, while in the queen and bishop you have only 2 piece types: linear and diagonal, with the diagonal type repeating once. This matters. It matters also in more complex groups of pieces, whether there is a queen or not. It is a separate feature from the respective piece values that you can not achieve by tuning values.

But I think we almost stole the very nice thread of Bram, I would enjoy a couple of more such beautiful games.

[beram]

Another match against a fresh Stockfish 1510 (Timestamp: 1381815361) against the new Komodo 6 resulted again in a clear Stockfish victory
This time with 57%

Code: Select all

Komodo 6 - Stockfish 1510, 4cpu, Blitz 5m+3s
AMD X6 1090T 3200 Mhz, (Fritz, 2013.10.15 - 2013.10.16)
                                  Score     12345678901234567890123456789012345678901234567890
-----------------------------------------------------------------------------------------------
 1: Stockfish 151013 64 SSE4.2  28.5 / 50   0101===1=10==1=1======111===100====011=0=1===10=1=  (+15 -8 =27)
 2: Komodo 6 64-bit             21.5 / 50   1010===0=01==0=0======000===011====100=1=0===01=0=  (+8 -15 =27)
-----------------------------------------------------------------------------------------------
50 games: +15 =27 -8

In this position Stockfish played a4 ! and created a very nice deep positional breakthrough, based on opposition of the king
Komodo 6 needs about 6 minutes till depth 34 on my system to see that a4 is best move
Stockfish 1510 sees this very fast as winning, Houdini 3 needs also minutes. Stockfish 4 needs 2 minutes.
Comments are welcome

[d] 8/5p2/4p1p1/pp1kP1Pp/2pP3P/P1P2P2/3K4/8 w - - 0 1

Code: Select all

2: Stockfish 151013 64 SSE4.2 - Komodo 6 64-bit, Komodo 6 -Stockfish 1510, Blitz 5m+3s 2013 8/5p2/4p1p1/ pp1kP1Pp/2pP3P/P1P2P2/3K4/8 w - - 0 1 Analysis by Komodo 6 64-bit 6cores: 38. Ke3 = (0.00) Depth: 20 00:00:02 197kN 38.Ke3 = (0.00) Depth: 21 00:00:02 339kN 38.Ke3 = (0.00) Depth: 22 00:00:02 636kN 38. Ke3 = (0.00) Depth: 23 00:00:02 782kN 38.Ke3 = (0.00) Depth: 24 00:00:02 1873kN 38.Ke3 = (0.00) Depth: 25 00:00:02 5056kN 38. Ke3 = (0.00) Depth: 26 00:00:02 6717kN 38.Ke3 = (0.00) Depth: 27 00:00:03 14486kN 38.Ke3 = (0.00) Depth: 28 00:00:04 20590kN 38.Ke3 = (0.00) Depth: 29 00:00:04 27081kN 38.Ke3 = (0.00) Depth: 30 00:00:08 55720kN 38.Ke3 = (0.00) Depth: 31 00:00:11 83200kN 38.Ke3 = (0.00) Depth: 32 00:00:23 227mN 38.Ke3 = (0.00) Depth: 33 00:00:33 362mN 38.Ke3 = (0.00) Depth: 34 00:00:39 403mN 38.a4 = (0.16 !) Depth: 34 00:04:08 2025mN 38.a4 +/= (0.63 !) Depth: 34 00:06:46 3393mN (Bram, Sommelsdijk 16.10.2013)

2: Stockfish 151013 64 SSE4.2 - Komodo 6 64-bit, Komodo 6 -Stockfish 1510, Blitz 5m+3s 2013
8/5p2/4p1p1/pp1kP1Pp/2pP3P/P1P2P2/3K4/8 w - - 0 1

Analysis by Houdini 3 x64 6cores:
  =  (0.00)   Depth: 26/45   00:00:02  11236kN
38.Ke3 Kc6 39.Kd2 Kd7 40.Ke2 Kc6 41.Kd2 Kd7 
  =  (0.00)   Depth: 27/49   00:00:03  21290kN
38.Ke3 Kc6 39.Kd2 Kd7 40.Ke2 Kc6 41.Kd2 Kd7 
  =  (0.00)   Depth: 28/49   00:00:04  37026kN
38.Ke3 Kc6 39.Kd2 Kd7 40.Ke2 Kc6 41.Kd2 Kd7 
  =  (0.00)   Depth: 29/53   00:00:07  65632kN
38.Ke3 Kc6 39.Kd2 Kd7 40.Ke2 Kc6 41.Kd2 Kd7 
  =  (0.00)   Depth: 30/62   00:00:12  126mN
38.Ke3 Kc6 39.Kd2 Kd7 40.Ke2 Kc6 41.Kd2 Kd7 
  =  (0.00)   Depth: 31/62   00:00:19  212mN
38.Ke3 Kc6 39.Kd2 Kd7 40.Ke2 Kc6 41.Kd2 Kd7 
  =  (0.00)   Depth: 32/64   00:00:36  414mN
38.Ke3 Kc6 39.Kd2 Kd7 40.Ke2 Kc6 41.Kd2 Kd7 
  =  (0.00)   Depth: 33/64   00:00:59  688mN
38.Ke3 Kc6 39.Kd2 Kd7 40.Ke2 Kc6 41.Kd2 Kd7 
  =  (0.00)   Depth: 34/65   00:01:43  1213mN
38.a4 
  =  (0.05 !)   Depth: 34/65   00:03:01  2197mN

(Bram, Sommelsdijk 16.10.2013)

2: Stockfish 151013 64 SSE4.2 - Komodo 6 64-bit, Komodo 6 -Stockfish 1510, Blitz 5m+3s 2013
8/5p2/4p1p1/pp1kP1Pp/2pP3P/P1P2P2/3K4/8 w - - 0 1

Analysis by Stockfish 4 64 SSE4.2 6cores:

38.Ke3 
  =  (0.00)   Depth: 32/4   00:00:01  9969kN
38.Ke3 
  =  (0.00)   Depth: 33/4   00:00:02  14351kN
38.Ke3 
  =  (0.00)   Depth: 34/4   00:00:04  34643kN
38.Ke3 
  =  (0.00)   Depth: 35/4   00:00:05  50694kN
38.Ke3 
  =  (0.00)   Depth: 36/4   00:00:06  62321kN
38.Ke3 
  =  (0.00)   Depth: 37/66   00:00:48  498mN
38.Ke3 
  =  (0.00)   Depth: 38/66   00:00:48  503mN
38.Ke3 
  =  (0.00)   Depth: 39/66   00:00:51  523mN
38.Ke3 
  =  (0.00)   Depth: 40/66   00:00:54  556mN
38.Ke3 
  =  (0.00)   Depth: 41/66   00:01:03  635mN
38.a4 bxa4 39.Kc1 a3 40.Kb1 a2+ 41.Kxa2 Kc6 42.Ka3 Kb7 43.f4 Ka7 44.d5 exd5 45.f5 
  =  (0.08 !)   Depth: 42/66   00:01:34  954mN
38.a4 bxa4 39.Kc1 a3 40.Kb1 a2+ 41.Kxa2 Kc6 42.Ka3 Kb6 43.Ka4 Ka6 44.f4 Kb6 45.d5 exd5 46.f5 gxf5 47.e6 fxe6 48.g6 
  =  (0.16 !)   Depth: 42/67   00:02:12  1280mN

(Bram, Sommelsdijk 16.10.2013)