Author: Dan Newman
Date: 11:30:47 07/20/98
Go up one level in this thread
On July 20, 1998 at 13:35:50, Bruce Moreland wrote: > >I would like to ask you to do another test, if you could, since you have both >versions available. > >Run your suite maybe 20 times, ten times with the old version, and ten times >with the new version, but interleave them so that you alternate between the two, >and do this via a batch file, so it is all as automatic as possible. > >What you should see is slow/fast/slow/fast/slow/fast etc. What I am a little >afraid you will see is slow/fast/fast/fast/fast/fast etc. > >bruce I think I've seen this priming-the-pump effect--though perhaps not as pronounced as a 4% change in node rate. Is it due to branch prediction being pre-set by the first run or some sort of alignment variation? One of the biggest effects I see is that I get a much larger node rate the longer I run. If I run a test for 10 s I might get 170 knps. If I run for 40 s 200 knps. It does level off of course. I'm using a P6/200, so I've attributed this effect to improving branch prediction as the code runs. Any ideas? -Dan.
This page took 0.04 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.