Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Test position ==> Aristarch 4.4 vs Tao 5.4 [from WinBoard Forum]

Author: Gian-Carlo Pascutto

Date: 16:00:20 08/07/02

Go up one level in this thread


On August 07, 2002 at 18:52:32, Gerd Isenberg wrote:

>On August 07, 2002 at 17:01:42, Gian-Carlo Pascutto wrote:
>
>>On August 07, 2002 at 16:52:40, Bas Hamstra wrote:
>>
>>>Vincent claims you can get 40% extra from 2 CPU's by just letting 2 processes
>>>search the same position and just and only share the hashtable. Of course you
>>>need to lock the shared hash, else "race conditions".
>>
>>You get less than 20% unless you throw out nullmove or make special
>>adjustments to make the algorithm work better (but your search overall
>>worse).
>>
>>--
>>GCP
>
>I don't tried it yet, i heard that each thread should use a different move
>ordering.

Only on ALL nodes. On CUT nodes, you want identical moveordering, which
can be a problem.

I spent 2 weeks with this scheme, determined the speedup, and wasn't very
happy, so I'm trying other things now.

--
GCP



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.