Author: Sune Fischer
Date: 00:41:39 10/04/03
Go up one level in this thread
On October 03, 2003 at 22:14:53, Robert Hyatt wrote: >>Your last statement is inconsistent with your "If efficiency continues to climb, >>it is unbounded." >> > >It isn't if you remain "in context". > >You picked an equation with a simple limit. I think Vincent only said the efficiency would rise, this doesn't imply it's unbounded. I too got the impression that this was what he meant though. >Vincent's "equation" has no known limit yet, because the obvious limit >for N processors is <= N, but he has (for years) claimed efficiency > N >for N processors. Since N is the controlling term in any equation based >on N, and since N is not a limit, the speedup (to me) appears to be >unbounded, plain and simple. Any speedup whatsoever that is >N means it >must be unbounded. >And we already have the >N claim in writing, many times. > >(not from me or anybody that knows what they are doing, however). Anyone who understands the basics of alpha-beta knows it's absurd. The funny thing is it can happen in practise, unfortunately it's nothing to be happy about, it just means that your serial search isn't "optimal". I guess a program using reversed move ordering would be quite capable of consitently producing >2 speedups? :) However Vincent did have an interesting point, that with N threads he can do N probes to the hash table 'in parallel'. Seems this would lower the latency relative to running just one thread on twice as fast a processor where the latency (counting clocks) would just go up. That factor has got to be peanuts relative to the parallel overhead though. -S.
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.