Author: George Tsavdaris
Date: 16:57:25 02/02/06
As an answer to: http://www.talkchess.com/forums/1/message.html?483948 >Chrilly said Hydra got 75-80% vs Rybka and I have no reason not to trust him on >such issues. Yes, but there is a problem on this: Suppose that Dr Donninger is right and Hydra on ONE single processor is winning against Rybka on one processor by 77.5% (the mean value of 0.75 and 0.80). Then Hydra on 32 processors, full power, i would expect it with a rather pessimistic guess, to gain at least 15% to 20% more in the match against Rybka from what single Hydra did (15 to 20% is used as a sum difference and not with the usual meaning of % gain). That means 92.5% to 97.5% performance. And that is with the best at the moment PC-engine right now, Rybka. And all that with a pessimistic guess of 15% to 20% gain from 1 to 32 processors. So against the old Fritz 8 for example we could have even 98% ++ performances. Sorry but i can't accept this and i controvert his statements.... In my opinion he just wants to keep the legent of Hydra alive but without games i think this is not possible..... Note that the 15 to 20% gain difference from Hydra(32) to Rybka(1)(i say again that 15 to 20% is used here as a sum difference from 77.5& and not with the usual meaning of % gain), corresponds to a +220 ELO difference of Hydra(32) to Hydra(1) and that is perfectly logical even pessimistic guess..... But even if the difference would be less, and Hydra(32) win 90% against Rybka, it is damn hard to believe. Hydra scored against a perhaps buggy Shredder 9, 69% so how i can believe the 90%+ scores.....Actually i can't believe in any way scores more than 77% of Hydra(32) against another top program. And of course for Hydra(1) i expect results to be _much_ less, actually losing. Another logic: Hydra(16 CPU) Vs Shredder(perhaps even buggy version) 5.5 - 2.5 = 69% Hydra(1 CPU) Vs Rybka = 75% to 80% Odd right.....?
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.