Author: Vincent Diepeveen
Date: 20:15:29 09/24/01
Go up one level in this thread
On September 24, 2001 at 22:49:02, Robert Hyatt wrote: >On September 24, 2001 at 21:49:52, Vincent Diepeveen wrote: > >> >>So you say kind of here that it is NEVER possible in your viewpoint to >>get a 2.0 speedup when only measuring search times? >> >>Please give a clear statement here. >> >>Yes or No. >> >> > > >OK. For an average case, which is made up of multiple positions, the answer is >"no" unconditionally. > >For a single test case here and there, the answer is _obviously_ "yes". > >_everybody_ has reported a >2 speedup for 2 cpus on a random position or two >out of a set. This is expected since we are dealing with randomness in the >move ordering and random can sometimes be good, sometimes bad. > >Look at my DTS numbers. at least one > 2, several well below 2. > >Average is a scientific-looking 1.9... > >If you report >2.0 speedups consistently, you are simply going to get >laughed at over and over. It is just not possible unless your serial >search is bad. So either your results (parallel search) are bad, or >your serial search is bad. In either case, "bad is bad". Define 'bad' here. 'bad' relatively compared to others who all use futility pruning, and who do recursive search, or 'bad' in absolute terms? I've said over and over again that i tried to improve my move ordering in all kind of ways. I have way more code than you have in crafty. If i order my moves like you do in crafty, then i search a ply less.
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.