Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Robert, a little question ...

Author: Vincent Diepeveen

Date: 15:05:06 09/20/00

Go up one level in this thread


On September 19, 2000 at 11:05:01, Frank Quisinsky wrote:

>On September 19, 2000 at 10:16:19, Robert Hyatt wrote:

>A questions to Crafty:
>I see that Crafty 17.13 is not faster then Crafty 17.11 but the new SMP version
>is faster.

>At the moment is the dual technic factor 1.8. Do you think that factor 1.9 or
>1.95 is possible ?

Diep gets 2.0 at 2 minutes a move , IF YOU HAVE AT LEAST 150MB HASH.

Note that diep is a slow searcher AND using multiprocessing AND each
process is SMP, where all parallel programs i know aren't SMP except
cray blitz. Not SMP: the process having the recursions can't abort
when the last moves searched are done by a different process. Apart
from that the chance that the first move gives a cutoff in a 'beta' node
is quite small in diep compared to most programs (yet it's using
very nodes a ply compared to other programs despite that it's not
forward pruning yet).

At levels of 30 minutes a move at a fast PIIIdual i see average speedup
of way over 2.0, more like 3.0.

You can theoretical be much much faster as 2.0. consider this.

Suppose you split in a position x. Now suppose that process 1 searches
a move A, and process 2 searches move B. Now move A is a long and heavy
search, but process 2 quickly gets a cutoff for move B.

So single cpu you would be hours busy with move A, but now you get a
quick cutoff for move B.

>I have interest to get more information why is not more then factor 1.8
>possible. I think a very interesting message for my News Ticker because more and
>more persons have interest to buy a dual system an I think this is a interesting
>question about SMP.

Where diep's speedup dual is great, i'm lacking a lot of tests at 4
processors for speedup. Yet it seems that getting 4.0 out of 4 processors
is nearly impossible (some forward pruning versions of diep could achieve
this but the versions without forward pruning did much worse). Dual
however is getting a great boost when at the newyear we will see the
first thunderbirds dual at ABIT dual motherboards. I can't
wait to get one of those!

>See here (little zip file, Crafty bench results on dual and single PIII):
>http://amateurschach.in-trier.de/schach/download/sonstige/crafty_bench2.zip

>Frank



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.