Fritz 11 = Rybka3 at what ply?

[Uri Blass]

Going by the depth that engines report or obey is really a meaningless exercise, as every engine measures depth differently. A meaningful question would be how much shorter time Rybka needs to reach an equivalent playing strength a Fritz 11.

This you can simply determine by playing a time-odds match, e.g. 40 moves/12 min, and specifying a time-odds factor of 12 for Rybka, so that it has only 1 min for 40 moves. (without ponder, of course). Keep increasing the time-odds factor as long a Rybka is still winning.

SInce you have thought about this, why don't you do it? It would be interesting to have such data, all run thru (say) BayesElo, to give a good estimate of 3:1 time odds (no pondering) is worth X +/- N Elo.

Actually, this might be something I could do on our cluster later on. It would definitely get rid of the old "doubling is 50 elo" or whatever rule-of-thumb that doesn't seem very accurate to me...

things may be dependent on the time control
It is possible that you get 80 elo for doubling at 1+1 time control and 50 elo for doubling at 60+60 time control.

More time is always important but I suspect that diminishing returns can be proved if you play many games.

Uri

[Erik Roggenburg]

Going by the depth that engines report or obey is really a meaningless exercise, as every engine measures depth differently. A meaningful question would be how much shorter time Rybka needs to reach an equivalent playing strength a Fritz 11.

This you can simply determine by playing a time-odds match, e.g. 40 moves/12 min, and specifying a time-odds factor of 12 for Rybka, so that it has only 1 min for 40 moves. (without ponder, of course). Keep increasing the time-odds factor as long a Rybka is still winning.

H.G. is absolutely right in this. I think testing could easily be done in any GUI. I'm most familiar with Fritz, and I know it can be done. Too bad my hardware sucks and I'm in the process of updating.