Author: Robert Hyatt
Date: 18:51:48 05/04/01
Go up one level in this thread
On May 04, 2001 at 17:57:36, Jesper Antonsson wrote: >On May 04, 2001 at 14:48:14, Robert Hyatt wrote: >>My 60M figure is "peak". To compare that to DB you have to use 1000M nodes >>per second. It would _still_ be a long way away. > >Well, what I remember is that they reported around 400M nodes peak and 200M >nodes average. Anyway, a factor of 4-16 is not something I consider very much, >it isn't more than two to six years of Moores Law. :-) However, it is still an >open question how good DB was at evaluation. Those guys were smart and could >throw silicon at the eval terms, so it's possible that they had significantly >better eval than state-of-the-art chess software of today. On the other hand, >it's possible they didn't. If you do the math: 480 chess processors, 1/2 at 20mhz, 1/2 at 24mhz, you get an average of 22mhz, which at 10 clocks per node means an average of 2.2M nodes per second per processor. Times 480 and you get 1 billion. Peak of course, but it _could_ reach that peak. Hsu claimed his search was about 20% efficient which would take that to roughly 200M... On a 64 cpu alpha it is _possible_ that Crafty might exceed 60M nodes per second. But in reality it would be searching like a 40M node per second sequential processor due to the .3 efficiency loss for each processor. Still, it would be _very_ fast. Just not as fast as deep blue by quite a ways... And then there is the evaluation problem. I _know_ I don't do in my eval what they did in theirs as I would probably be another factor of 3-5 slower if I did... > >However, in six years, chess software should have progressed somewhat, there >will probably be six man EGTB, much more RAM for hash tables than DB had, and an >architecture that allows more efficient parallel search than DBs. Thus I think >it will be possible to have a DB strength chess machine on a general purpose >high end server in six years from now. I don't think we will be able to do / use all the 6 man egtbs within 6 years. The size of all of those will be mind-boggling. We are approaching 100 gigs and have yet to do anything but 3 vs 3 with no pawns... > >How well does your SMP algorithm scale, by the way, if you don't mind me asking? >64 processors are quite a lot. > >Jesper Antonsson If the _hardware_ scales well, then the SMP algorithm will do fine. My rough estimate for speedup is N-(N-1)*.3 for a rough speedup estimate. Or to make it simpler, .7N where N is the number of processors... 64 nodes should scream... I will have some data by the Summer I hope...
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.