Stockfish 1.8 live for the IPON list ...

[IWB]

... here:

http://www.inwoba.de

Have fun
Ingo

[Tomcass]

TEST 1.
120 GAMES
Quad 2,33 (All engines 4 cores)
Book: HS Masterbook 2.0
GUI: Fritz 12
Ponder: Off
Time control: 5 min +0

2010-07Stockfish1.8JA 2010

1 2 3 4
1 IvanHoe 57a4 JR 10.0 - 10.012.0 - 8.014.0 - 6.0** 36.0/60
2 Deep Rybka 4 w32 10.0 - 10.010.0 - 10.011.5 - 8.5 ** 31.5/60
3 Houdini 1.02 w32 4_CPU 8.0 - 12.010.0 - 10.011.5 - 8.5 ** 29.5/60
4 Stockfish 1.8 JA (x4) 6.0 - 14.08.5 - 11.58.5 - 11.5 ** 23.0/60

TEST 2 .
120 GAMES

(This computer is about three times faster than the previous quad).

i7 975 (8 threads per engine, although Stockfish 1.8 used only 5)
Book: HS Masterbook 1.0
Gui: Fritz 12
Ponder: Off
Time control: 5 min +0



201007Stockfish1-1 2010

1 2 3 4
1 IvanHoe 57a4 (x8) 10.5 - 9.510.0 - 10.012.5 - 7.5** 33.0/60
2 Deep Rybka 4 x64 9.5 - 10.512.0 - 8.010.0 - 10.0 ** 31.5/60
3 Houdini 1.02 x64 POPCNT 8_CPU 10.0 - 10.08.0 - 12.010.0 - 10.0 ** 28.0/60
4 Stockfish 1.8 JA 64bit (x8) 7.5 - 12.510.0 - 10.010.0 - 10.0 ** 27.5/60


GLOBAL
240 GAMES

1 IvanHoe 57a4 JR 69.0/120
2 Deep Rybka 4 63,0/120
3 Houdini 1.02 57.5/120
4 Stockfish 1.8 JA 50,5/120

My feelings:
- Houdini performed in these tests below than usual.
- Stockfish 1.8 JA is only a little bit better than 1.7. It performs better, the faster the computer and/or the longer the time control. The same happens, as all of us know very well, with Deep Rybka 4.
- IvanHoe 57a4 JR is probably the best engine today at fast time controls.

Regards,

Tom

[Frank Quisinsky]

Hi Tom,

short to your statement to Rybka 4 ...
An other opinion here ...

Rybka 3 / 4 not important.

Rybka scored with longer time controls not better. Possible Houdini others from Ipp fam. scored better as Rybka with shorter time controls. But vs. other engines Rybka 3 / 4 scored with longer time controls not better as the "most" other engines.

After a test I made with x games on my machines.

Rybka 3:

Game in 5 = in combination with 20 other programs (each one vs. each one a lot of games) Rybka 3 scored around 55 ELO stronger as SF 1.6.3 with 32-bit.

40 moves in 10 minutes = again in combination with 20 other engines (each one vs. each one) Rybka 3 scored around 35 ELO stronger as SF 1.63 ... same if I compare it with others engines.

40 moves in 20 minutes (so far only 500 games for each engine) ... Rybka scored around 25 ELO stronger as SF 1.6.3.

After all I do and test ... Rybka 3 (I think 4 too, not sure) lost points vs. other strong engines with longer time controls.

If you test Ipp. fam. engines you will not see it because I think its clear that Ipp fam. engines are stronger as Rybka 3 / 4 with very short time controls.

Best
Frank

Other example:
I believe with 40 in 40 on actual fast hardware SF 1.7.1 JA x64 have the same level as Rybka 4, perhaps Rybka is 10 ELO stronger ... a very little bit stronger.

But same opinion to your other opinion to SF.

[Tomcass]

Hi Frank,

Thank you very much for your comments. You are right when you assume that I test mainly Ippo engines, although crossed with commercials such as Rybka and Naum.

I am sure that your opinion is better substantiated than mine for diferent reasons:

1.- Your tests are made in a more scientific way than mine. (For instance I use HS Masterbooks versions 1.0 and 2.0 with no limit of moves).
2.- You have tested for years all the best available commercial engines. (Although I have bought all of them, I prefer to test the continuous flow of new Ippolit sons developed by an impressive team of excellent programmers. By the way, I would like you to give a look to the new Ivan55b. It seems very promising... )

For this reason I always start my statement with words like 'My feelings are' or similar.

Anyway it is very interesting for me to know and learn from your experienced vision about all these topics, Frank.

Regards from Barcelona,

Tom.

[Frank Quisinsky]

Hi Tom,

the problem is, that we don't have results with longer time controls. Means really good results with a lot of games.

Possible all what I wrote is nonsens and Rybka will get with 40 in 120 a booster.

Look:
Hiarcs 12.1 is a good example:
Hiarcs 12.1 is clearly stronger in blitz as all other "for me known and tested" engines. Around 20-30 ELO more in fast blitz as 40 in 10 time control. Now the question: How many ELO lost Hiarcs 12.1 with tournament time control to the others, possible that Hiarcs get the booster with such a time control.

To the IPP fam. engines:
I don't have interest to test the engines. But each one can used the SWCR databases for add IPP fam. engines.

I have a bigger interest to add in the next months more amateurs in my SWCR-32 and SWCR-64.

Have fun with Ivan55b

Best
Frank

PS: I tested for months Firebird 1.0 Beta in "game in 5 - Athlon x2 4600, socket 939 with ponder " and "40 in 10 on Intel Quad Core Q9550, 2.83GHz with ponder".

The results are:
In Blitz after 500 games: +37 to Rybka 3
40 in 10 after 720 games: +23 to Rybka 3

The only engine of Ipp fam. I test a little bit without to public the results. Need a lot of time for do it because I try in these times different of these engines. All have ponder problems and crashed from time to time. Not very interesting for the automatically Shredder tournaments I play. I am on 3 days in the week not at home and after a crash the tournament stopped.

[IWB]

Hello

Stockfish 1.8 finished the first part of the games with a moderate win of Elo, but well within error margin, compared to the previous version.

http://www.inwoba.de

Bye
Ingo