Author: Robert Hyatt
Date: 14:52:24 06/20/02
Go up one level in this thread
On June 20, 2002 at 16:13:14, Tom Kerrigan wrote: >On June 20, 2002 at 15:58:46, Robert Hyatt wrote: > >>Did you remember the discussion here a few years back when Bruce and I took >>our NPS on P6/200 machines, and then on equal alphas, that my program scaled >>much better than his because of the 64 bit operations helping me while he >>had none in his program. > >No, I don't remember that discussion. This is the only pertinent and interesting >thing you've said in the entire thread; I'm baffled as to why you waited so long >to say it. > >What do you mean by "much better"? > >And I'm not convinced Bruce's program was relatively slower because of the >64-bit operations. There are other differences between your programs. > >-Tom So? The _main_ difference is that everything he does is 32 bit. the majority of what I do is 64 bit. That is the main difference. I don't remember the actual speedups. jason ran Crafty (I think this was the Paris WMCCC event). DEC loaned us a plain 500mhz alpha. Bruce had made arrangements to use one of the Cryo-cooled alphas at 767mhz or some such number. At some point he ran crafty and ferret both on the cryo machine for testing, and all I remember is that for Crafty, alpha/P6 produced a larger number than alpha/P6 for Ferret. IE I believe that on a 500mhz alpha, my number was > 3 (alpha NPS / P6 NPS) was > 3. Perhaps 3.5 but I really don't want to say that was the number since I didn't see it myself. Ferret produced a 2.5 I believe, assuming 3-3.5 was my number. The only conclusion there had to be that the 64 bit operations were saving me more time than they were for Ferret, which is no surprise since it didn't have any 64 bit ops at all except for possibly hashing. The difference was definitely significant, even if not a factor of two or more.
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.