Author: Robert Hyatt
Date: 07:44:52 08/19/03
Go up one level in this thread
On August 18, 2003 at 12:10:08, Russell Reagan wrote: >On August 18, 2003 at 11:39:08, Robert Hyatt wrote: > >>Anything is possible, of course. But you won't just lose "some >>speed". You will lose a _bunch_. This is a well-understood problem. > >Getting a bit more off topic here, but you could achieve an Nx speedup on N >processors using min-max, right? Yes. I did this in my Ph.D. dissertation as one test case. Linear speedup on minimax is not hard. > With pure min-max this wouldn't work well in >practice, but maybe some kind of selective min-max could be competitve with >alpha-beta based algorithsm as SMP/HT machines become more and more affordable. >The selective min-max could still make use of alpha-beta/pvs/mtd/etc. > The problem is the exponential property. With alpha/beta you can search _twice_ as deep as with minimax alone. No way to catch up with parallel minimax. >My thinking is that this wouldn't pay off great dividends immediately, but >eventually we will probably run into some kind of barrier as to new improvements >to a single cpu's speed. One obvious continuation is for the cpu creators to >start piling single cpus together as they've already started doing with HT. If >in 30/40/50+ years we have a cpu that has a million cpus within it, we might see >the payoffs there.
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.