Author: Robert Hyatt
Date: 07:56:07 07/16/04
Go up one level in this thread
On July 16, 2004 at 08:59:42, Tord Romstad wrote: >Which influence does the branching factor of a two-player, zero-sum game have >on the efficiency of a parallel search? Is the speedup when using multiple >processors independent of the branching factor, or does the speedup increase >(decrease?) when the branching factor increases? Intuitively I would expect >the speedup to increase with the branching factor, but I have no experience >with parallel search. Basically no effect. IE you can do just as well with a non-null-move non-selective-search program as you can with a program that does both. The issue is move ordering more than anything else, because when you split the search at some point, you want to split where you need to search all moves, rather than at a point where you expect or get a fail high. > >In case it is not clear, when talking about "branching factor" above, I am not >talking about what is commonly called the "effective branching factor" (which is >typically in the range 2-3 for a chess program using recursive null move >pruning), >but about the average number of legal moves (around 30-40 in a typical chess >middle game). > >Tord
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.