Author: Adam Oellermann
Date: 05:34:51 09/05/01
Go up one level in this thread
On September 05, 2001 at 07:53:50, guy haworth wrote: > >In the Ken-Thompson-themed ICGA Journal (June, 2001), Ernst Heinz published his >latest self-play experiment results. > >Engines with different guaranteed-depth(?)-parameters were pitted against each >other. > >The matches of the experimetn (3,000 games each) suggest that: > > 12-ply was 84 ELO points better than 11 ply > 11-ply was 92 ELO points better than 10 ply > 10-ply was 115 ELO points better than 9 ply > >Fairly strong indications of decreasing returns from increasing search. No >doubt a proper statistical analysis will follow. > > >An extra ply seems to require 4-6 times the 'effective power', so a factor of 36 >- if realised across the system - is only 2 plies. > >On this basis, an engine might be some 160 ELO points better on hardware giving >a power factor increase of 36. > > >G I guess the juicy part will not be just running a naive search two ply deeper, but just imagine the heavy-duty stuff you could do in the eval with 70GHz and still reach current NPS levels. I would suggest that depth increases of more than 2 ply are to be expected because of reduced branching factor from complicated eval stuff as well as better extensions in tactically alive positions. I expect that people who currently have an advantage from implementing in assembly will lose that edge - it'll just be too expensive to keep implementing ever more complex eval/search logic in ASM. As a result, I accept the argument about diminishing returns to depth, but I do not agree with the notion of diminishing returns to hardware improvements - mainly because heavier hardware lets you do heavier chess-specific knowledge while still staying efficient. I would be very interested to see some of the CCC illuminati's predictions as to what the NPS of leading chess programs will be 5 years hence. For myself, I predict that they won't go too far past 2m NPS, but they'll make it count a lot more. I also hope that as a result anti-anti-computer chess will improve to the point where a computer is a reasonable substitute for a strong human in training. - Adam
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.