Author: scott farrell
Date: 20:17:09 12/01/03
Go up one level in this thread
On December 01, 2003 at 13:18:02, Tord Romstad wrote: >On December 01, 2003 at 08:33:56, scott farrell wrote: > >>On December 01, 2003 at 06:57:13, Tord Romstad wrote: >> >>>From the initial position, I compute the attack tables about a million >>>times per second with this code. From WAC1, the rate drops to about >>>700,000 times per second. The high nodes/second count of Rebel (which >>>has similar tables) makes me believe that it is possible to do this >>>many times faster. >>I am also working on incremental attack tables. >>I also found huge slow downs. > >I think you misunderstand. I have never used incrementally updated >attack tables, only attack tables generated from scratch at every >node. And I have no idea how much they slow me down, because they >have been in my program since the beginning. my mistake > >>I am not going to ditch them just yet. I am going to see if I can put the extra >>info to good effect. > >Yes, it is very useful several places in the evaluation function. I >use them to detect mate threats, hanging pieces, compute king safety, >mobility, space, centre control and so on. There is hardly any >component of my evaluation function which does not make extensive >use of attack tables. > >>I think if you can get extra knowledge from the extra work, then you have to not >>use it. >> >>I would be real happy with anything near 1mill/sec - in java I get only about >>200 Knps on an athlon 2Ghz > >I don't get anything near 1 million nodes per second. I meant that with >my new code I can generate the attack tables 1 million times per second >from the opening position. Even with my code doing it "incrementally" and hence faster I think, I generate all attackFrom and attackTo tables. I still only get 350K nodes generate per second during perft from the opening position on a 2Ghz athlone machine. > >Actually, my engine searches less than 200 Knps in most middlegame >positions on a PIV 2.4 GHz. I expect the new engine I am working on >to be even slower when it is finished. > >Tord
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.