Author: Jason Deines
Date: 13:45:41 11/06/97
Go up one level in this thread
On November 06, 1997 at 11:39:47, Robert Hyatt wrote:
>On November 06, 1997 at 07:30:56, Peter Herttrich wrote:
>
>>Hello Bob, Hello Crafty-Freaks,
>>
>>i took a look yesterday evening into the game Crafty-Fritz
>>in the 2nd round.
>>
>>And there i detected a strange thing.
>>I cannot reproduce the move 13. Rd1 for Crafty.
>>I switched on the annotate:
>>********
>>[Event "1997.10.26"]
>>[Site "Paris"]
>>[Date "1997.10.27"]
>>[Round "2"]
>>[White "Crafty / Alpha 500"]
>>[WhiteElo ""]
>>[Black "Fritz / P2-300"]
>>[BlackElo ""]
>>[Result "1/2-1/2"]
>>[Annotator "Crafty v13.10"]
>>{annotating only white moves.}
>>{using a scoring margin of +0.000 pawns.}
>>{search time limit is 10:00}
>>
>> 1. d4 d5
>> 2. c4 c6
>> 3. Nf3 Nf6
>> 4. Nc3 dxc4
>> 5. a4 Bf5
>> 6. e3 e6
>> 7. Bxc4 Bb4
>> 8. O-O Nbd7
>> 9. Qb3 a5
>> 10. Nh4 Bg6
>> 11. g3 Qb6
>> 12. Nxg6 hxg6
>> 13. Rd1
>> {12:-0.510 Rd1 g5 f3 Qc7 Na2 Bd6 Rb1 O-O Bf1 Nb6 Bd2
>> Nbd5}
>> {12:-0.234 Na2 Bd6 Qxb6 Nxb6 Bb3 g5 f3 g4 Bd2 Nbd5 Nc3
>> Nxc3 bxc3 g5 fxg4 Nxg4}
>
>I don't have a clue here. I assume you are using the 13.10 source code
>from my ftp machine? If so, this is *exactly* what we played with for
>all
>10 rounds.
>
>We did use more hash tables than you, I believe our machine had 256M of
>memory
>and we used 192M for hash. That machine was also at least 2.5X faster
>than yours, but you let it search far longer than we could, so that
>shouldn't
>matter.
>
>I will try this on my machine as well, but perhaps we did have some sort
>of
>bug that was unknown...
>
Here's some more information on how I configured crafty in Paris. The
machine did indeed have 256MB of RAM, but I found that if I exceeded
128MB total memory usage, performance suffered, more than offsetting any
gain from the larger tables. I suspect it is some sort of memory-bus
quirk specific to the somewhat older system I was using. The hash
settings I wound up using were 96M general, 20M pawn.
Some other things to keep in mind are learned values that might have
effected this position, and with the line you posted from the endgame,
that the Paris version had a subset of the 5-man tablebases available.
If you are only testing with the 4-man (or none at all) that will
undoubtably affect the eval.
Before I left I ran some tests to confirm that the values generated from
Crafty/Alpha were the same as from Crafty/Intel, and they were. For the
curious, the specific machine configuration was an Alphastation 500:
Alpha 21164/500MHz, 256MB RAM, 2MB L3 cache, 4GB SCSI hard disk, NT 4.0
w/ service pack 3, VC++/Risc(Alpha) 5.0 w/ service release 2.
Finally, it would be most illuminating to have the original logfiles
from the Paris machine, but unfortunately the floppy I had them on got
trashed somehow and I have been unable to recover any information off of
it.
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.