Author: Robert Hyatt
Date: 20:13:33 11/19/02
Go up one level in this thread
On November 19, 2002 at 17:07:41, Gian-Carlo Pascutto wrote: >On November 19, 2002 at 16:48:55, Robert Hyatt wrote: > >>That is a 40% improvement if I did my math right. Which is in line >>with what most knowledgable people claim a bitmap program should do since not >>all instructions need 64 bits of data (branches, loop counters, a few others). > >If I remember correctly, the 'speedup' of Crafty when running >on 64-bit hardware vs. the one of normal programs wasn't remotely >close to 40%. (I'm basing this on SPEC results) How can you use SPEC to conclude anything about whether crafty does better on 64 vs 32 bits? There is only one chess program in SPEC. Comparing a chess program to an integer program (not chess) is meaningless. Most of the integer spec programs deal with lots of memory streaming... unlike chess engines. That is why Crafty was selected for SPEC in the first place, it is a _different_ kind of integer application. > >It's been observed that different pro's got different speedups >when SSDF switched from AMD K6 450 to Athlon 1200's as well. Certainly. But they are running the _same_ executable. That's not the same as natively compliing for each machine so that the only variable is the program and data structures. The SSDF is subject to other issues such as poor instruction scheduling, poor instruction choices for different machines, etc... > >The design of the program can benefit or get a penality from >a new platform by other effects than just the board representation. If the program exhibits certain types of behavior. But, again, two chess engines, compiled for the same two machines, using the same compiler, takes a lot of that out of the equation... > >-- >GCP
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.