Author: Angrim
Date: 02:00:21 12/10/99
Go up one level in this thread
On December 10, 1999 at 00:06:05, blass uri wrote: >I am interested in this question because I understood that >64 bit signature may be counter productive at correspondence time control >and I would like programs to avoid mistakes even in 24 hours/move. > >How much slower are program if you use 96 or 128 bit signature instead of 64 bit >signature? > >Uri maybe 1-5% slower, the real reason I use 64 bits is space rather than speed. According to my math, the odds of a 64 bit hash key giveing a false match is absurdly low. Assumeing a 10meg entry hash table which would take 240meg for my hash entry size, 200k hash probes per second, (many more nodes than probes, unless you probe hash in the qsearch) and 24 hours per turn, the odds of haveing a single false positive in that turn's search should be slightly less than 1% (10,000,000*200,000*60*60*24)/(2^64)=.0093 If your program crashes when it gets a false match, then this would be mildly worrying since your program would crash for no good reason every few years when running full time on very fast systems, but usually a false match will have little or no impact on which move is played. Angrim(useing 64 bit hash key myself)
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.