CEGT - rating listsSeptember 11th 2011

[Werner]

Hi all,

our actual rating lists are online and can be found under the attached links.

40 / 20:
New games: 2953; 52 different engines
Total: 534.721

NEW Engines

33 Komodo 3.0 x64: 3144 - 661 games (+30 to previous version)
49 Komodo 3.0 w32: 3106 - 400 games (+32 to previous version)
539 Dirty 08112011 x64: 2683 - 414 games (good start here in our list)
645 Cheng3 v1.04 x64: 2610 - 302 games (+85 to version 1.03)
670 Arasan 13.1 x64 1CPU: 2595 - 207 games (startrating)

UPDATES
6 Critter 1.2 x64 4CPU: 3241 - 2304 games (-1)
8 Critter 1.2 x64 2CPU: 3221 - 600 games (+5)

40 / 4:
New games: 10.500
All games now: 877.639

New Engines

31 Komodo 3.0 x64: 3160 - 1700 games (+22 to prev. version)
65 Komodo 3.0 w32: 3067 - 1700 games (+27 to prev. version)
216 Spark 1.0 w32 1CPU: 2916 - 1200 games (+12 to prev. version)
510 Redqueen 1.0.0 x64 1CPU: 2702 - 1300 games (+60 to previous version!)
Arasan 13.1 x64 1CPU: 2615 - 1000 games (+20 to previous version)
685 Cheng3 v1.04 x64: 2612 - 1000 games (+200 to older version 1.02b)

Some Updates
4 Critter 1.2 x64 4CPU: 3270 - 1700 games (-1)
4 Critter 1.2 w32 1CPU: 3163 - 1500 games (-9)
44 Stockfish 2.1.1 w32 1CPU: 3119 - 1800 games (-4)
374 Bobcat 2.75 x64 4CPU: 2815 - 1000 games (-14)
330 Equinox 0.96y w32 1CPU: 2839 - 1700 games (-8)

40/120
See here our new single-list:
http://cegt.siteboard.eu/f4t18-new-40-1 ... -list.html
We are now testing Komodo 2.03 x64.

A big „Thank you“ to all testers as usual!!

Links

40/20: http://www.husvankempen.de/nunn/rating.htm
Blitz: http://www.husvankempen.de/nunn/blitz.htm
40/120: http://www.husvankempen.de/nunn/rating120.htm
Tester: http://www.husvankempen.de/nunn/testers/testers.htm
Elo-comparison: http://www.husvankempen.de/nunn/Replay/ ... arison.htm
Games of the week: http://www.husvankempen.de/nunn/40_40%2 ... on/gow.jpg

Werner Schuele
CEGT-Team

[lkaufman]

What percentage (roughly) of your tests on Komodo 3 were run on machines that could use the sse version, and if they could use it did they always do so? It seems to make a bigger difference on Komodo than on other programs.

Thanks.

[Werner]

Hi Lary,
I used sse (80%) of the games - and of course the opponents on this pc too if available.

[IWB]

Hello Werner,

Hi Lary,
I used sse (80%) of the games - and of course the opponents on this pc too if available.

I wrote this to the CCRL as well, so it isw just a cut and paste:

I dont think that mixing engine versions which are definately different is the right way for a rating list! What about other engines?

Houdini
Komodo 80%
Rybka
Critter
Umko
...

And how do you control that all engines which are available in SSE and NONSSE get the same ratio of games?

I know that the difference is bigger, but thinking this to the logical end you could combine 32 and 64 bit versions as well.

Bye
Ingo

[Werner]

I saw a speed difference of about 10% here on my pc and I don´t think we can measure such an elo difference with our tests.
Maybe other things like opening positions make more differences in the tests.

best wishes
Werner

[IWB]

I saw a speed difference of about 10% here on my pc and I don´t think we can measure such an elo difference with our tests.
Maybe other things like opening positions make more differences in the tests.

best wishes
Werner

I dont htink that similar "speed" is a argument. These versions ARE different, they sometimes produce even different MV (Check K2.03 in SSE and NON SSE, it is most obvious there).
Then why 10%. Would you still throw it together at 15, 25 or 35%. Where is the border line and why exactly there?
All this is a principle question. But I agree that adapting hardware or books might cause more distortion than that 10%. It is just that books or hardware ar concesions to a more diverse test enviroment, while mixing SSE and nonSSE versions simply is no necesssity! They could easily be split into two version, alone the CEGT has to track that ..

Bye
Ingo

[lkaufman]

I saw a speed difference of about 10% here on my pc and I don´t think we can measure such an elo difference with our tests.
Maybe other things like opening positions make more differences in the tests.

best wishes
Werner

Without taking any position on the question of how large the speed difference should be to justify separate tests, I just thought I'd mention that we expect the gap between SSE and non-SSE to be even larger with our next Komodo release.