Author: Tony Werten
Date: 02:00:17 06/20/02
Go up one level in this thread
On June 20, 2002 at 03:49:14, Gian-Carlo Pascutto wrote: >On June 20, 2002 at 03:37:13, Tony Werten wrote: > >>On June 20, 2002 at 03:14:53, Kurt Utzinger wrote: >> >>>Personally spoken, I prefer matches with ponder=off. And this due to an >>>experiment made some years ago even played on_two_PC 486/86 with_two_programs at >>>level 120 min/40 moves. The first match over 20 games was played with ponder=on, >>>the second match [same openings] with ponder=off. And you will hardly believe >>>it, but in the match with ponder=off the overall average search depth per move >>>was higher than in the match with ponder=on. This seems to confirm what Dieter >>>Buerssner wrote on this subject. He said that he would prefer to do his private >>>tests with ponder=off as this method would use CPU time more efficiently because >>>the CPU cycles for pondering on wrong moves are not wasted. Maybe someone should >>>repeat such a test with the latest programs to have a good comparison between >>>the various programs. >> >>I don't think you need tests. >> >>Suppose we have 100 cycles to spend for 2 moves. ( 1 own, 1 ponder ) >> >>with pondering on: >> >>you get 25 cycles for the first move, 25 for the second. Assuming pondering is >>correct 75% of the time you get 25+(0.75*25)=43.75 cycles effectively. >> >>with pondering off: >> >>you get 50 cycles for the first move, 0 for the second=50 cycles effectively >> >>Tony > >These maths only hold on a single machine, but Kurt's test was run on two... > Oops. Yes, this was for 2 engines,1 processor. Tony >-- >GCP
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.