Author: Dan Newman
Date: 04:43:58 10/11/00
Go up one level in this thread
On October 10, 2000 at 12:16:33, Steve Maughan wrote: >Dan, > >Your hash hit rate is way higher than mine (only 4.4%). I'm using a two hash >table system with a depth based store criterial and an always replace criteria. >What the general consensus - is this too simplistic? Is a multiple hit >attempts algorithm much better? > That's what I'm using (the two level table). I believe Vincent has said that he uses a multiple hit scheme to good effect. I've always worried a bit about the memory traffic though, but if you don't probe in the qsearch, it turns out (at least in my case) to not be very bad at all. (I only got a small change of a few percent in node rate when going from a 100 MHz to 133 MHz memory clock. And the node rate scales almost linearly with processor speed up to 933 MHz.) I may try multiple re-hashing some day, but it's a low priority since the two-level seems to work so well... >My first move cutoff ratio is also poor at 85%. I assume the two are linked! > As an experiment I tried zeroing out my hash table move, so that it couldn't be used. The first move cutoff ratio dropped from 92.6% to 89.8%, so it didn't have an enormous effect. But then 89.8 is getting close to 85... -Dan. >Many thanks, > >Steve > >>Mine (Shrike) is doing this in just under 3s (1.27M nodes) on a P3/600. >>It's getting about a 35% hit rate on the hash table, so this may well >>be the problem. Those hash table moves are really important in getting >>good move ordering (and a low branching factor). >> >>I also keep stats on the fraction of cutoffs from the first move tried >>and the fraction that are from hash table moves. In this case I'm >>getting 92% of the cutoffs from the first move and 25% from the hash table.
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.