Author: Robert Hyatt
Date: 10:48:04 06/22/99
Go up one level in this thread
On June 22, 1999 at 12:11:26, James Swafford wrote: >On June 22, 1999 at 05:43:15, Vincent Diepeveen wrote: > >>On June 22, 1999 at 02:27:32, Jouni Uski wrote: >> >>>On June 22, 1999 at 02:20:06, Mark Longridge wrote: >>> >>>>Does anyone know if there are any programs aside from crafty which >>>>support more than 1 cpu? I'm interested in multi-cpu software, >>>>both shareware and commercial. >>>> >>>>Thanks >>> >>>Fritz5.32 supports 2 prosessors under Windows NT. So You can run engine-engine >>>with ponder! >>> >>>Jouni >> >>DIEP: n processors, where n is any number > 0. >>email me: diep@xs4all.nl >> >>DIEP runs both under linux and NT. >> >>Speedup out of 4 processors at world championship was 4.0 > >How do you get a speedup of 4.0 with 4 processors? >Do you claim *no* loss? For n processors, is the speedup >always = n? > >I've heard others state that 3.2 or 3.3 is a good speedup >for 4 processors.... > >Please enlighten me. > >-- >James > > >> >>The parallellism in DIEP is an improved version of the Cray Blitz >>parallellism. >> >>Many thanks to Hyatt for explaining me how to do parallellism and >>giving me the opportunity to run and test at 4 processors! >> >>Greetings, >>Vincent Until his program is fully debugged, there is little point in speculating. once it is, it is easy to do a full 10-ply search with 1 cpu, and then a full 10-ply search with 4 cpus and see what happens. Cray Blitz got fairly close to 4 (I don't remember the exact numbers and am not at my office to check the DTS article in JICCA). Crafty is currently hitting 3.2-3.3 on an average of several test suite positions.... I don't see how to hit 4.0 without perfect move ordering, or absolutely worst -first move ordering. Either of those is trivial to go 4.0x with, but with the normal 90-95% right ordering we see, 4.0 is _very_ difficult... At least for me...
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.