Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Crafty 16.7

Author: Robert Hyatt

Date: 12:13:58 06/02/99

Go up one level in this thread


On June 01, 1999 at 20:30:36, Gregor Overney wrote:

>It is a combination of the things: code changes (such as in evaluate.c), number
>of processors involved, choice of OS, choice of compiler etc.
>
>It appears like searching a global minimum in a configuration space with lots of
>tempting local minima. One slight change, and the search may be unsuccessful.
>
>FEN 8/k7/3p4/p2P1p2/P2P1P2/8/8/K7 w - - 0 1 works on some people's Linux box
>using mt=2 and 16.8 (ply = 22) and shows up much later under NT 4 (using mt=2
>and 16.8).
>
>But it is an interesting example to proof the point that "things" on NT and
>LINUX are handled differently enough that they cause different results for
>identical input and source code. Is it VC++ or just the way NT is designed?
>
>Gregor


Not _just_ NT vs Linux.  I get significant variability on the same hardware,
the same position, the same everything. Because the parallel search is highly
non-deterministic.  NT isn't the problem at all.  IT seems to do the SMP stuff
quite well in fact...



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.