Author: Robert Hyatt
Date: 06:44:46 08/30/99
Go up one level in this thread
On August 30, 1999 at 02:26:34, blass uri wrote: >On August 29, 1999 at 21:20:43, Robert Hyatt wrote: > >>On August 29, 1999 at 18:17:13, Mogens Larsen wrote: >> >>>On August 29, 1999 at 15:22:41, Robert Hyatt wrote: >>> >>>>Ed doesn't either. And I wouldn't be surprised if everyone else doesn't spend >>>>a lot of time on ponder=off games either. It is simply 'unnatural' to run a >>>>program that way... and most of us would rather spend time tuning the program >>>>in the state it will play games, not in some crippled state that a user might >>>>use to play games. IE do we also tune for (a) tiny transposition tables; (b) >>>>no opening book; (c) no databases (endgame); (d) modified user parameter >>>>settings; (e) any other random thing a user might try??? >>>> >>>>IE I do my testing in the configuration that plays the best/strongest. Not in >>>>configurations that someone might use "just because it is there..." >>> >>>I've been following the discussion with great interest and I have a couple of >>>questions, mostly due to ignorance. >>> >>>If you play an engine-engine match on one computer with permanent brain on and a >>>match with permanent brain off. What match would most likely be the best >>>estimate of the difference in strength? What are the complications with >>>permanent brain? Some suggest that it's the same for both, but there might be a >>>difference prioritywise concerning processortime, or? >>> >>>Best wishes... >>>Mogens >> >> >>Neither, unfortunately. Here's why. >> >>Assume one null-move program and one non-null-move program. If you use ponder >>mode, both will get 1/2 the machine basically. Which means that in essence, >>the programs will be running on machines 1/2 the speed of the computer you are >>using. That hurts a null-mover more, because reduced depth allows some critical >>null-move failures that deeper depths 'fix'. So there, you get skewed results. > >I do not think that there is a rule that null movers earn more from time >relative to non null movers and it may be depend in the program. >I have no proof that Junior earn less from time relative to Null movers. > >Uri I didn't give it as a "rule" I gave it as an actual "observation". Thorsten saw this when he moved Crafty from a P5/90 or whatever onto a P6/200 type machine (I think he actually had a K6 but am not sure). I used to see significant null-move problems when I ran on a P5/133 a few years ago... and crafty lost many games, commonly when the opponent could get a pawn or bishop at f3/f6 and his queen could get to h3/h6 for an unstoppable mate thread. Null move hid this. When I moved to the P6/200, I didn't see this _nearly_ as often. And when I went to the quad P6 and now the quad xeon, it happens even less. The problem is that a null move can completely crush a 5-6 ply search, as at ply=1 the program grabs a pawn, at ply=2 the opponent plays Qh6, at ply=3 we try a null-move and that hides everything about the unstoppable mate threat and the program concludes that all is well. Add a couple of plies and that doesn't happen at the root..
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.