New Rating Lists (FGRL)

[fastgm]

Hi!

My new website with 4 different Rating Lists is online.
On aim is to see how develops the playing strength with increasing time control at exact same conditions.

http://www.fastgm.de

Andreas

[Adam Hair]

Thanks for sharing this, Andreas!

Would it be possible for you to include ratings calculated by Ordo?

[PaulieD]

Thank you!

Will you be including Stockfish 4 soon?

[fastgm]

Of course, Ordo ratings are included.

[Laskos]

Hi!

My new website with 4 different Rating Lists is online.
On aim is to see how develops the playing strength with increasing time control at exact same conditions.

http://www.fastgm.de

Andreas

Very interesting! Just looking by, I saw one noteworthy thing: the NPS from your AMD to i7 machines vary from 78% to 106%. I didn't know engines could behave so differently on different machines, that's a lot.

[fastgm]

Unfortunately not, because it lasts several weeks to finish the first gauntlet at time control 240+2.40 sec.

[Laskos]

Hi!

My new website with 4 different Rating Lists is online.
On aim is to see how develops the playing strength with increasing time control at exact same conditions.

http://www.fastgm.de

Andreas

Thanks, excellent lists. The factor is always 4 between time controls, and it's easy to visualize the scaling of engines, as on logarithmic scale the points are evenly spaced. I use your Ordo ratings with average 3000 to plot the ratings at different TC (viewed as logarithm in basis 10 of the base time on horizontal axis) for the stronger 5 engines.


Average being 3000, all engines will tend to this value as TC goes to infinity, so the approach towards this value is understandable as rating compression, but it also clear that there are differences. Houdini and especially Bouquet (an Ippo derivative) seem to scale worse than Komodo and especially Stockfish, with Rybka somewhere in the middle. With the arrival of Stockfish 4, it seems that the odds of winning nTCEC are pretty evenly divided between Houdini, Stockfish and Komodo, if the extrapolation from your ratings to longer time controls is valid.

[velmarin]

Hi!
My new website with 4 different Rating Lists is online.
Andreas

Oh, yes.
A pretty and practical list.
Thanks.

Houdini and especially Bouquet (an Ippo derivative) seem to scale worse than Komodo and especially Stockfish,...
it seems that the odds of winning nTCEC are pretty evenly divided between Houdini, Stockfish and Komodo,...

Yes, you is right on the deficiencies of Bouquet, the version "Bouquet 1.6" I had to change several things.
The version on TCEC has all the "time control" remodeled, waiting makes good use of the "two hours" of the game.

Still, we agree, Houdini, Komodo and Stockfish are above.