Author: Robert Hyatt
Date: 06:17:54 06/29/00
Go up one level in this thread
On June 29, 2000 at 00:10:36, blass uri wrote: >On June 28, 2000 at 23:58:59, Robert Hyatt wrote: > >>On June 28, 2000 at 12:48:10, ujecrh wrote: >> >>> >>>I am still exploring some ideas and before doing anything that is already known >>>to be useless I prefer asking here if it is worth it: >>> >>>Some programs are known to be good blitz players and other seem to prefer longer >>>time controls. One of the reasons for it might be that the amount of knowledge >>>in the evaluation. I wonder if anybody did (or does) try to use more or less >>>knowledge depending on the time controls ? >>> >>>Same idea for hardware speed. Recent posts stated that some new strong engines >>>might be weaker on slow hardware than older ones. Many programs evaluate >>>computer speed when started, do they tweak their evaluation according to this ? >>> >>>(I am just talking here about eval and not search. The fact that some search >>>techniques benefit more than others of faster machines has already been >>>discussed here) >>> >>> >>>Ujecrh >> >> >>I would suggest two points: >> >>(1) the evaluation for a slow machine is not necessarily the same evaluation >>I would write for a fast machine. The fast machine will be aware of more >>tactical things, relieving the eval of noticing some things. >> >>(2) the search for a slow machine is not necessarily the same search I would >>write for a fast machine. >> >>In short, the program for a fast machine is not necessarily the same as a >>program for slower hardware. > >I think that the program can calculate a number (nodes per move) in the >begininng of the game and decide about the evaluation and the search rules based >on this number. > >Uri I think that (a) tuning it would be a monumental task; and (b) doing this would make it slower and way more complex to write, debug and test, And (c) is the real killer: who has access to all the various machines that the program might run on, so that it can be tuned? It takes years to write and tune an engine to run well on current hardware. The complexity added to try to make it dynamically adjust to hardware speed would be _huge_. And who really cares how a current program runs on an 80C88 or MCC6800 or a 6502?
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.