Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: GNUChess takes it to the hoop against Crafty

Author: Robert Hyatt

Date: 14:11:19 03/03/05

Go up one level in this thread


On March 03, 2005 at 13:29:01, Matthew Hull wrote:

>On March 03, 2005 at 13:14:27, Dann Corbit wrote:
>
>>On March 03, 2005 at 09:41:14, Matthew Hull wrote:
>>
>>>On March 03, 2005 at 08:49:08, Andrew Wagner wrote:
>>>
>>>>This is an impressive result. Did you tinker with GNUChess' code at all? What
>>>>opening book are you using? It didn't look like it out-booked crafty. Any
>>>>thoughts on why it performed so well?
>>>
>>>
>>>This GNUChess is stock.  The book is also the stock GNUDhess book (which took
>>>all night to process from the 160,000 pgn file, BTW).
>>>
>>>The only tweak I did was to implement GCC PGO profiling.  This mormally breaks,
>>>but for some reason, it works on my old PIII 500mhz desktop (RH9).  So I built
>>>it with PGO on the old machine and now run it on the less-old machine.  :)
>>>
>>>Perhaps someone who has studied GNUChess 5.07 can contribute more on it's level
>>>of sophistication.  It seems to be basically sound,  but not a particularly deep
>>>searcher.  A 64 bit machine might give it a significant speedup, since it is a
>>>bitboard program.
>>>
>>>To be fair, I've suspected that Crafty has some kind of "hole" in it somewhere,
>>>since 19.15.  It plays the endgames better than it has ever played them, but
>>>sometimes it loses it's way in the middlegame.  Just an "impression" from
>>>watching many games it's played on ICC.
>>
>>I would not take too much from two games at high speed.
>>
>>If GnuChess wins in a tournament of 30 games, then it is time to raise the
>>eyebrows.
>
>
>I agree completely.  But it is rather lucky that this account has played crafty
>on ICC exactly twice with a non-insignificant hardware disadvantage.

You can never tell about the hardware.  My dual xeon does not sit idle during
the day.  I am generally doing something on it most of the time, which means the
ICC crafty doesn't get the entire box.  Often it gets 1 CPU out of 2, but with
my SMP search, that is way worse than just using one thread to use one cpu...

No idea when the games were played and whether I was doing anything serious or
not, but if someone wants to play a match, I can make certain that the machine
is not doing something else at some specific time...



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.