Author: Dave Gomboc
Date: 08:23:31 04/15/99
Go up one level in this thread
On April 15, 1999 at 09:36:51, Bruce Moreland wrote: > >On April 15, 1999 at 06:27:26, Ulrich Tuerke wrote: > >>On April 15, 1999 at 05:30:55, Rick Andrews wrote: >> >>>About how much weaker do chess programs play when playing against >>>each other on the same computer? >>>Thanks, Rick Andrews >> >>Assumed that when playing on the same computer/processor, pondering will be >>disabled.Also assumed that programs are doing essentially the same when >>pondering: expect the next move of the last PV and search. >>Pondering enabled can give theoretically a factor of 2 in performance when all >>guesses are hits (you have your and the opponent's time). In this theoretical >>extreme case pondering could make a difference of about 40 - 50 ELO points. >>In practice, there are less hits. So I arbitrarily guess that the difference >>could be 20 - 30 ELO points. > >A normal Windows program does its initialization stuff, then sits in a message >loop, waiting for stuff to happen. While it is sitting in the message loop it >is blocked, meaning that it is not using CPU. > >This model doesn't fit chess programs very well. They want to be doing >something all of the time, and the classic implementation involves a recursive >function, so you are off in this enormous call stack doing stuff all of the >time. > >It's possible to make this thing its own thread, or you can call a message loop >occasionally while searching. > >If "pondering" is turned off, the program should just sit there in the main >message loop between moves. I don't think that all of the programs do this. I >remember testing on a single machine a long time ago, and seeing NPS much >reduced even when the other program was sitting there doing nothing. > >Does anyone know of a program that for sure uses CPU even when it is not >thinking? > >bruce CM 4000, CM 5000, and CM5500 were all reported to exhibit poor "idle" behavior, i.e. eating cycles despite not being on the move and having pondering off. CM6000 has not yet been criticized in the same way, but nobody's said that they've fixed it either. <shrug> Dave Gomboc
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.