Author: Robert Hyatt
Date: 13:55:11 06/13/02
Go up one level in this thread
On June 13, 2002 at 14:03:19, Dann Corbit wrote: >On June 13, 2002 at 10:43:42, Robert Henry Durrett wrote: > >>An idea: >> >>Perhaps it is possible to know, in advance, which computer SSDF will use for >>testing of a particular chess program. >> >>If so, then do you suppose it would be possible to get a higher rating by >>optimizing the code for that particular computer? >> >>How much higher? > >5 ELO tops. Maybe 10% improvement in speed if they spent a great deal of effort >writing special tricks for a particular CPU. In fact, it would be an incredibly >stupid waste of effort. Far better would be to spend the same time trying to >improve their fundamental algorithms. I disagree. Look at what the fritz guys did to Chess Tiger on the chess server with the h4 opening. If a program _knew_ its opponent, it could certainly use that to good effect. That would have been a +200 point SSDF rating swing. I don't say that a single program can cook every other one, but that was a well- known cook that worked very well. game after game. Without an opening book for either side after h4. Actually it might have been something like 1. e4 h5 where black won every game, I don't recall since it didn't affect my program, and since I don't try to cook my book for other programs even though on the chess server I do know exactly who I am playing.
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.