Deep Rybka 4 Vs Fire 1.2/1.3

[JWorcester]

150 Games 75 positions played alternately as white/black. All the positions have material imbalances either piece/exchange/pawn sacrifices.

Played on a dual core E8400 O/C to 4Ghz. 256Mb Hash each and ponder off. TC is 1m+1s.

1 Deep Rybka 4 w32 +37/=82/-31 52.00% 78.0/150
2 FireBird 1.2 w32 +31/=82/-37 48.00% 72.0/150

Equates to a +14 Elo Performance.

Currently after 98 games between DR4 and Fire 1.3 the score is 51-47 to Fire which equates to a +14 Elo performance by Fire 1.3. I'll update when the match is finished but DR4 doesn't seem to blow Fire 1.3 or 1.2 away.

[JWorcester]

Games completed now.

1 Fire 1.3 w32 +50/=65/-35 55.00% 82.5/150
2 Deep Rybka 4 w32 +35/=65/-50 45.00% 67.5/150

Result equates to a +35 performance for Fire 1.3 compared for DR4.

[Carlos777]

Interesting results. Thanks Justin!

Regards,
Carlos

[ozziejoe]

this idea of playing with material imbalances is very cool i think. I would love to get ahold of the positions?

I have one qeustion. Does such a test set magnify elo differences? e.g., it does not include more drawish positins that commonly occur in tournament chess.

[JWorcester]

this idea of playing with material imbalances is very cool i think. I would love to get ahold of the positions?

I have one qeustion. Does such a test set magnify elo differences? e.g., it does not include more drawish positins that commonly occur in tournament chess.

I only have them as a database of PGNs at present. If you PM me your email I'll happily send them to u.

I don't know if it magnifies ELO differences. E.g.

1 Minutes/Game + 1 Seconds/Move 2010

1 Stockfish 1.7 JA +31/-28/=43 51.47% 52.5/102
2 Rybka 3 32-bit +28/-31/=43 48.53% 49.5/102

Only 51 positions but statistically suggesting that Rybka 3 was very similar to SF 1.7 in strength which is in keeping with what others are finding. However as I do watch some of these games I'd rather see

[d]r1qk3r/pbnpb3/1p3pQ1/2pn4/8/P5P1/1P2PPBP/RNBR2K1 b - - 0 17
Radjabov-Leko Morelia/Linares 2008

[d]rn1q1rk1/p3ppbp/1p2b1p1/8/2pp3B/2P1PB2/P2N1PPP/1R1QK2R w K - 0 13
Moiseenko-Svidler Sochi 2005

than games that start with a drawish position.

I've played similar matches between various engines. The positions vary from match to match simply because I'm always tinkering with which positions I include. Sometimes I think I've found a position that is imbalanced materially but the lines are forcing and resolve quickly into a materially balanced drawish position and if this holds true for several engine matches then I delete that game. My results are broadly in line with what I see on rating lists that include large numbers of games but then again I don't pretend to have a statistically robust list.

[Leto]

Just finished a private match between Deep Rybka 4 x64 4CPU and Fire 1.3 x64 4CPU, using the same settings I use for my CEGT testing. Q9650 at 3.8ghz, 4gb ram, 128mb hash, perfect 8.32, 40 moves in 3 minutes repeating (adapted to CEGT 40/4).

Deep Rybka 4 x64 4CPU (3265) scored 30.0-20.0 +18 -8 =24 60% +70 elo performance.

Fastest decisive game in this match:

[Event "Rybka 4 x64 4CPU testing 7"]
[Site "Microsoft"]
[Date "2010.05.30"]
[Round "47.1"]
[White "Deep Rybka 4 x64 4CPU"]
[Black "Fire 1.3 x64 4CPU"]
[Result "1-0"]
[ECO "D46"]
[Annotator "0.56;0.40"]
[PlyCount "77"]
[EventDate "2010.05.29"]
[EventType "tourn"]
[TimeControl "40/180:0/0:0/0"]

{Intel(R) Core(TM)2 Quad CPU Q9650 @ 3.00GHzW=14.1 ply; 488kN/s; Perfect
v8.32.ctg B=18.2 ply; 6,988kN/s; Perfect v8.32.ctg} 1. Nf3 {B/0 0} d5 {B/0 0}
2. d4 {B/0 0} Nf6 {B/0 0} 3. c4 {B/0 0} e6 {B/0 0} 4. Nc3 {B/0 0} c6 {B/0 0} 5.
e3 {B/0 0} Nbd7 {B/0 0} 6. Bd3 {B/0 0} Bb4 {B/0 0} 7. a3 {B/0 0} Bxc3+ {B/0 0}
8. bxc3 {B/0 0} Ne4 {B/0 0} 9. O-O {B/0 0} Nxc3 {B/0 0} 10. Qc2 {B/0 0} dxc4 {
B/0 0} 11. Bxc4 {0.56/16 4} Nd5 {0.40/20 5} 12. e4 {0.54/16 5} Ne7 {0.39/19 6}
13. a4 {0.56/15 4} a5 {(0-0) 0.38/19 7} 14. Ba3 {(Bg5) 0.56/14 8} O-O {0.42/19
6} 15. Ne5 {(Bd3) 0.63/15 23} Re8 {0.40/18 7} 16. Rad1 {0.63/15 5} Nb6 {(Nf8)
0.48/17 12} 17. Ba2 {(Be2) 0.75/14 5} Nd7 {(f6) 0.62/17 6} 18. f4 {0.85/14 4}
Nxe5 {0.77/18 6} 19. dxe5 {0.85/14 2} Qc7 {0.95/19 5} 20. Bd6 {0.97/14 3} Qb6+
{1.09/19 26} 21. Rf2 {(Kh1) 0.97/15 6} h6 {1.03/18 12} 22. Rb1 {(Kh1) 1.17/13 1
} Qd8 {1.03/18 4} 23. Rd2 {(f5) 1.21/15 3} g5 {(Ng6) 0.99/18 8} 24. fxg5 {(f5)
1.70/14 5} hxg5 {1.90/18 16} 25. Rf2 {1.97/14 9} b6 {(Ng6) 2.13/17 10} 26. Qe2
{(Qd2) 2.62/13 5} Ng6 {3.02/16 4} 27. Rbf1 {(Qh5) 3.84/13 4} Re7 {(Ra7) 3.18/
17 6} 28. Qh5 {4.93/12 2} Raa7 {3.02/15 0} 29. Rf3 {(Qxg5) 5.05/13 1} g4 {3.78/
15 5} 30. Qxg4 {5.05/12 1} Qe8 {(Kg7) 4.58/15 4} 31. Qh5 {(Rh3) 6.82/12 2} Kg7
{(Qf8) 7.05/15 3} 32. Rh3 {8.65/10 1} Qh8 {6.97/13 0} 33. Qg5 {8.91/12 0} Re8 {
9.21/15 3} 34. Qf6+ {(Rxh8) 13.16/12 1} Kg8 {0.01/0 0} 35. Rxh8+ {13.16/13 1}
Nxh8 {0.01/0 0} 36. Qg5+ {13.16/13 1} Kh7 {12.32/19 3} 37. Rf3 {13.16/13 23}
Ng6 {12.32/18 0} 38. Rh3+ {13.16/12 3} Kg8 {16.30/20 6} 39. Qh6 {(Qh5) 13.16/
13 18} 1-0