Author: Uri Blass
Date: 03:16:48 09/21/01
Go up one level in this thread
On September 21, 2001 at 00:02:14, Robert Hyatt wrote: >On September 20, 2001 at 23:05:00, Slater Wold wrote: > >>On September 20, 2001 at 10:16:10, Robert Hyatt wrote: >> >>>On September 20, 2001 at 00:51:25, Slater Wold wrote: >>> >>>>Same position. Different hash. >>>> >>>>Just wanted to see how fast it took to solve, so I did not let it complete. >>>> >>>>Same *exact* machine. Only thing changed, was the hash in the crafty.rc file. >>>> >>> >>> >>> >>>Test is no good. _never_ use SMP to compare different things. If you want >>>to run using two processors, you need to run each position a few dozen times >>>and take the _average_ time as the result. SMP has way too much variablility >>>in search time to compare with anything. >>> >>>I _never_ use it when debugging, for example. >> >>Interesting. Uri had said something about it, and so I did it just on a whim. >>Wasn't trying to be *exact*. >> >>I will test it on just 1 CPU again, and see what happens. >> >>Bob, do you think that more hash will result in quicker solution? >> >> >>Slate >>> > > >Yes. I ran such a test for "Komputer Korner" several years ago. For >"middlegame" positions, a hash size that is _way_ too small resulted in >search times about 2x longer than those produced when the hash size was >appropriate. There is a point beyond which larger hash doesn't help at >all, of course. But this depends on the speed of your hardware and the >speed of the program you are using and exactly what is hashed. > >Crafty is less bothered by smaller hash tables since it doesn't hash in the >q-search. But a factor of 2x when going from just right to way too small is >reasonable for middlegames... The relevant positions were endgames and not middle game and this is the reason that I thought that a difference in hash tables can explain the difference between slater's results and Dann corbit's result. Uri
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.