Author: Robert Hyatt
Date: 08:39:08 08/18/03
Go up one level in this thread
On August 17, 2003 at 23:16:19, Uri Blass wrote: >On August 17, 2003 at 22:15:53, Robert Hyatt wrote: > >>On August 17, 2003 at 03:37:22, Johan de Koning wrote: >> >>>On August 16, 2003 at 05:13:23, Gian-Carlo Pascutto wrote: >>> >>>>On August 16, 2003 at 03:24:47, Johan de Koning wrote: >>>> >>>>>>Nondeterminism is something you can live with. Forget about getting >>>>>>the engine working on multiprocessor if you don't. >>>>> >>>>>So far you sound like someone trying to mimic Vincent. :-) >>>> >>>>But you sound like someone who avoided my point :) >>>> >>>>At some point you must choose between determinism and performance. >>>>I don't want to go to lengths to keep the engine deterministic only >>>>to find out later it can't be maintained AND I've lost time and >>>>speed trying to put off the inevitable. >>> >>>I'm not avoiding your point, I'm simply denying it. :-) >>> >>>There isn't such a thing as the point of no determinism, there are many points >>>at which choices can be made. If performance gain is small (playing games) I >>>will prefer determinism and simplicity. If gain is large (interactive analysis) >>>I will prefer determinism and perfection. In case determinism is impossible >>>(deep) I will prefer to maintain determinism for all other cases. >>> >>>>Hence the Santa Claus reference. You can stubbornly keep believing, >>>>but one day, he'll stop bringing presents. Or maybe it was just me who >>>>was a bad boy. >>> >>>If you quit being a bad boy Santa will return. :-) >>> >>>... Johan >> >> >>If you ever do a parallel search, forget about determinism. If you don't, >>forget about winning games. :) > >one cpu has chances not only to win games but also to win tournaments. > >A lot of programs won tournaments inspite of having only one cpu. So? People win the lottery each week too. At extremely high odds against it happening. But for _consistency_ the faster you go, the better you will do. And a parallel search is clearly faster. > >The king did it in the dutch championship when more than one cpu of some >opponents did not help them compute "probability". > >Same for Shredder in the past when Ferret,Junior,Fritz had more than one cpu. > >When computers become faster I expect the advantage of 2 cpus against 1 cpu to >be less important and things like better order of moves may be more important. > >It is also possible to be deterministic with more than one cpu. >You may lose some speed relative to being not deterministic but still do >significantly better than one cpu. > >Uri Anything is possible, of course. But you won't just lose "some speed". You will lose a _bunch_. This is a well-understood problem.
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.