Author: Vincent Diepeveen
Date: 08:01:25 12/20/02
Go up one level in this thread
On December 19, 2002 at 16:50:32, Dann Corbit wrote: >Suppose that we are searching at long time control, and the hash table rapidly >fills up. This can throw move ordering in the toilet, if we can't find >transpositions any more. > >What sort of replacement scheme is best for very long searches (several >minutes)? Of course also interesting is to draw conclusions regarding how many probes you should try. at today p4s you get 128 bytes at the same time. So with 8 probes of 16 bytes you can get in general within 1 or 2 sequential cache lines for free 8 entries versus 1 entry. Of course 8 probes at big search depths completely outperforms a very slow 2 table approach. updating is very easy. just use a value that adds the number of moves in the game done to the search dept left. that's how i do it in DIEP. Of course i also increment that counter when i do takebacks. basically it gets incremented with each search. Best regards, Vincent
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.