Author: Brian Richardson
Date: 09:30:53 06/15/00
Go up one level in this thread
On June 13, 2000 at 12:18:34, TEERAPONG TOVIRAT wrote: >Hi, > >I have 3 questions. >1. Does two level transpositional table have significant >advantages over one level table ? If so,how much we >gain in % ? > <snip> > >Thanks, >Teerapong I have tried one and two level tables in Tinker every which way I could think of. One level has never been slower, and sometimes up to 10% faster. This is with 512K or more _entries_ (not total MB hashtable size in memory). My suggestion would be to get a one level table working first. The store and probe code should be very localized. Once that is working, try a two level implementation. Results in your program may be very different. Note that Tinker also does not maintain a hash ID, which would get bumped for each new search. Instead, I "age" hash drafts after the computers move and again after the players response (after pondering) when the predicted move is not made. Even for 1M entries, this does not seem to take any significant amount of time. I have also tried the ID approach with both 1 and 2 level tables. I suspect the older ICCA articles favoring 2 level tables were based on much smaller numbers of entries and what would now be older hardware. As I recall, there was even an ICCA piece mentioning that for larger sizes the differences were very small. Brian
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.