Author: Dan Newman
Date: 10:19:10 09/06/00
Go up one level in this thread
On September 06, 2000 at 11:06:42, Philipp Claßen wrote: >Hi! > >I want to add a extra transposition table for quiescence search to my own >program. But I´m not sure how to avoid that important entries are overwritten by >less important ones. In a normal search the distance is a good criterion but in >a quiescence search? Perhaps the current ply? > >Phil I've tried this a couple of times, but the main effect seemed to be to slow down the node rate. I don't recall whether it dropped the scores on testsuites, but it certainly didn't improve them--or I would have kept it. I don't recall what I did about depth. I think I might have stored them all at a depth of zero. I probably didn't have a very good implementation of this, and in a different program it might work fine, so you should probably try it anyway... [Also, I just stored into my regular hash table and didn't try a separate one.] Whether it works probably depends on how much work you are already doing at a qnode and how large your memory bandwidth is. The more work you are doing, the more this can save. The higher the bandwidth, the less it costs. I suspect in my case that the massive increase in hits to the memory outweighs any benefit. -Dan.
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.