Critter with SSE and without
Posted: Sat Jul 09, 2011 7:33 pm
Hello
I run a little test with Critter 1.2 x64 with and without SSE42.
The speed difference on my AMD Test Hardware is about 2.8% which means somewhere between 2 and 3 Elo:
Running 2500 identical games for both versions it turns out that the result is nearly identical:
Critter 1.2 NOSSE got 1870 points, Critter 1.2 SSE42 1863 points. It is hard to see more than statistical noise in 7 points difference after 2500 games.
Bayeselo has 1 Point difference in favor for the NON SSE version, Elostat made 2 Elo out of this.
I personaly would argue that, with such a small difference, it is better to release just one version to avoid confusion about what to run where!
For better comparision I will remove the SSE42 Version from the IPON.
Bye
Ingo
I run a little test with Critter 1.2 x64 with and without SSE42.
The speed difference on my AMD Test Hardware is about 2.8% which means somewhere between 2 and 3 Elo:
Code: Select all
Engine: Critter 1.2 NOSSE42 (256 MB)
by Richard Vida
16/30 0:01 +0.22 1.e4 e5 2.Nf3 Nc6 3.Nc3 Nf6 4.Bc4 Bd6
5.O-O O-O 6.d3 Na5 7.Bb3 Nxb3 8.axb3 b6
9.Be3 Bb7 10.Qe2 (2.407.717) 1453
...
23/48 4:31 +0.16 1.e4 e5 2.Nf3 Nc6 3.Bc4 Nf6 4.d3 Bc5
5.Nc3 O-O 6.O-O d6 7.Be3 Bxe3 8.fxe3 Be6
9.Nd5 Bxd5 10.Bxd5 Nxd5 11.exd5 Ne7
12.c4 f6 13.Qd2 Qd7 14.e4 (437.361.968) 1609
best move: e2-e4 time: 5:01.594 min n/s: 1.609.816 nodes: 485.359.616
Engine: Critter 1.2 SSE42 (256 MB)
by Richard Vida
16/30 0:01 +0.22 1.e4 e5 2.Nf3 Nc6 3.Nc3 Nf6 4.Bc4 Bd6
5.O-O O-O 6.d3 Na5 7.Bb3 Nxb3 8.axb3 b6
9.Be3 Bb7 10.Qe2 (2.407.717) 1510
...
23/48 4:23 +0.16 1.e4 e5 2.Nf3 Nc6 3.Bc4 Nf6 4.d3 Bc5
5.Nc3 O-O 6.O-O d6 7.Be3 Bxe3 8.fxe3 Be6
9.Nd5 Bxd5 10.Bxd5 Nxd5 11.exd5 Ne7
12.c4 f6 13.Qd2 Qd7 14.e4 (437.361.968) 1657
best move: e2-e4 time: 5:01.515 min n/s: 1.657.942 nodes: 498.860.032Code: Select all
Critter 1.2 NOSSE42 : 2500 (+1433,=874,-193), 74.8 %
Deep Fritz 12 32b : 100 (+ 61,= 38,- 1), 80.0 %
Zappa Mexico II : 100 (+ 75,= 21,- 4), 85.5 %
Toga II 1.4 beta5c BB : 100 (+ 70,= 25,- 5), 82.5 %
Naum 4.2 : 100 (+ 43,= 48,- 9), 67.0 %
Loop 13.6/2007 : 100 (+ 80,= 20,- 0), 90.0 %
Deep Rybka 4 : 100 (+ 21,= 62,- 17), 52.0 %
Deep Onno 1-2-70 : 100 (+ 75,= 23,- 2), 86.5 %
Deep Shredder 12 : 100 (+ 47,= 46,- 7), 70.0 %
Crafty 23.3 JA : 100 (+ 85,= 15,- 0), 92.5 %
Jonny 4.00 32b : 100 (+ 85,= 14,- 1), 92.0 %
Deep Sjeng c't 2010 32b : 100 (+ 55,= 38,- 7), 74.0 %
Umko 1.1 SSE42 : 100 (+ 70,= 27,- 3), 83.5 %
spark-1.0 SSE42 : 100 (+ 56,= 38,- 6), 75.0 %
Houdini 1.5a : 100 (+ 22,= 47,- 31), 45.5 %
HIARCS 13.2 MP 32b : 100 (+ 65,= 33,- 2), 81.5 %
Protector 1.4.0 x64 : 100 (+ 67,= 29,- 4), 81.5 %
Strelka 2.0 B : 100 (+ 71,= 27,- 2), 84.5 %
Spike 1.4 32b : 100 (+ 53,= 37,- 10), 71.5 %
Gull 1.2 : 100 (+ 60,= 36,- 4), 78.0 %
Deep Rybka 4.1 SSE42 : 100 (+ 20,= 51,- 29), 45.5 %
Deep Junior 12.5 : 100 (+ 67,= 27,- 6), 80.5 %
Stockfish 2.1.1 JA : 100 (+ 22,= 59,- 19), 51.5 %
Komodo 2.03 DC SSE42 : 100 (+ 29,= 52,- 19), 55.0 %
Hannibal 1.1 : 100 (+ 58,= 38,- 4), 77.0 %
Umko 1.2 SSE42 : 100 (+ 76,= 23,- 1), 87.5 %Code: Select all
Critter 1.2 SSE42 : 2500 (+1441,=844,-215), 74.5 %
Deep Fritz 12 32b : 100 (+ 71,= 25,- 4), 83.5 %
Zappa Mexico II : 100 (+ 70,= 27,- 3), 83.5 %
Toga II 1.4 beta5c BB : 100 (+ 78,= 22,- 0), 89.0 %
Naum 4.2 : 100 (+ 48,= 48,- 4), 72.0 %
Loop 13.6/2007 : 100 (+ 79,= 17,- 4), 87.5 %
Deep Rybka 4 : 100 (+ 21,= 54,- 25), 48.0 %
Deep Onno 1-2-70 : 100 (+ 77,= 21,- 2), 87.5 %
Deep Shredder 12 : 100 (+ 47,= 47,- 6), 70.5 %
Crafty 23.3 JA : 100 (+ 87,= 13,- 0), 93.5 %
Jonny 4.00 32b : 100 (+ 84,= 11,- 5), 89.5 %
Deep Sjeng c't 2010 32b : 100 (+ 54,= 39,- 7), 73.5 %
Umko 1.1 SSE42 : 100 (+ 78,= 21,- 1), 88.5 %
spark-1.0 SSE42 : 100 (+ 63,= 34,- 3), 80.0 %
Houdini 1.5a : 100 (+ 15,= 45,- 40), 37.5 %
HIARCS 13.2 MP 32b : 100 (+ 53,= 40,- 7), 73.0 %
Protector 1.4.0 x64 : 100 (+ 53,= 43,- 4), 74.5 %
Strelka 2.0 B : 100 (+ 73,= 25,- 2), 85.5 %
Spike 1.4 32b : 100 (+ 59,= 34,- 7), 76.0 %
Gull 1.2 : 100 (+ 57,= 35,- 8), 74.5 %
Deep Rybka 4.1 SSE42 : 100 (+ 23,= 56,- 21), 51.0 %
Deep Junior 12.5 : 100 (+ 63,= 33,- 4), 79.5 %
Stockfish 2.1.1 JA : 100 (+ 26,= 57,- 17), 54.5 %
Komodo 2.03 DC SSE42 : 100 (+ 26,= 42,- 32), 47.0 %
Hannibal 1.1 : 100 (+ 59,= 34,- 7), 76.0 %
Umko 1.2 SSE42 : 100 (+ 77,= 21,- 2), 87.5 %Critter 1.2 NOSSE got 1870 points, Critter 1.2 SSE42 1863 points. It is hard to see more than statistical noise in 7 points difference after 2500 games.
Bayeselo has 1 Point difference in favor for the NON SSE version, Elostat made 2 Elo out of this.
I personaly would argue that, with such a small difference, it is better to release just one version to avoid confusion about what to run where!
For better comparision I will remove the SSE42 Version from the IPON.
Bye
Ingo