Author: Howard Exner
Date: 13:52:06 11/09/98
Go up one level in this thread
On November 09, 1998 at 15:36:12, Ed Schröder wrote: >>Machine is a K6-233 in Dos mode and 60 MB hash. >>Combination is off and all else at default (the default mode for Rebel 10 when >>playing games). > >>Initially it plays g5 (2:37) but without the correct followup in the eval, >>namely d4. However it finds >>d4 (4:47 - 15th ply) when that position is reached. > >>These are from the game Kotov Vs Botvinnik: > >>8/8/4b1p1/2Bp3p/5P1P/1pK1Pk2/8/8 b - - id end08 - Ecm98; bm g5; >>8/8/4b3/2Bp2Pp/7P/1pK1Pk2/8/8 b - - id end09 - Ecm98; bm d4; > >>One puzzling thing is that I double checked these positions in Windows95 mode >>with 28 MB hash. The d4 move was not found in the 15th or 16th ply. I thought it >>was due to Windows but it turns out that the 60 Mb vs 28 Mb of hash somehow >>makes a difference. > >Yes, this can happen. In fact it happens a lot especially in end-games. > >Rebel (since Rebel8) uses the hash table also for other things. The rule is the >more memory for hash tables the better Rebel will play. So your observation >is a correct one. > >On the other don't be too surprised if you occasionally see these things happen >the other way around as I have seen the exceptional examples Rebel playing the >good move with 28 Mb and not with 60 Mb. It's not uncommon just a negative >side effect of a good algorithm. As long as they remain exceptions I can live >with it. Various threads in the past have reported similar findings for other programs. This could explain why some moves cannot be reproduced by the same program on different machines, as they might have different hash size settings. Anyway, it was nice that the new Rebel finds this classic sequence of moves from Botvinnik. That position has always intriqued me. The theme of one bishop holding off the passed pawns while the two widely seperated pawns (referred to by Botvinnik as "trousers") are unstoppable has come in handy on a few of my games.
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.