Author: Vincent Diepeveen
Date: 16:09:56 08/29/01
Go up one level in this thread
On August 29, 2001 at 09:33:13, Pham Minh Tri wrote: >On August 29, 2001 at 08:41:42, Vincent Diepeveen wrote: > >>On August 29, 2001 at 03:39:52, Pham Minh Tri wrote: >> >>>I see that recently many computer chess programmers prefer Athlon to Intel. I >>>have not any Athlon and experience about it (but planing to buy one). However, I >>>see the benchmark of TSCP (in Tom Kerrigan site) shows that Intel is faster than >>>Athlon in almost all cases. So I do not understand that preference: because of >>>price only? Or is TSCP not a typical chess program (in terms of complexity)? Or >>>do I miss something? >>>An additional question: Could VC6 optimize for Athlon? >>>Thanks. >> >>www.aceshardware.com to see how an intel optimized version of diep >>is already faster at K7. >> 120k nps at dual K7 1.2Ghz MP DDR ram >> 106k nps at a dual Xeon 1.7 Ghz intel RDRAM >> >>VC6 with sp5 + processor pack is considerably faster on AMD. >> >>Still trying to figure out here how much as i can't compile right now >>at the AMD. >> >>Best regards, >>Vincent > >I have read that page about Diep benchmark. Thanks for your information. I am >still amazing how your nps is so small, comparing with others. > >A little worry: does a _typical_ chess program have the same result? Have you >ever tried Crafty? (Perhaps your program is a special case because of too heavy >knowledge, isn't it?). Well i first thought that DIEP on paper would look worse on a P4 than crafty, but as tests reveal this is not the case. In general over the years it has been proven, with exception of 64 bits machines, that most strong chessprograms give about the same results. The exception being assembly programs. Like fritz is seemingly very fast on a P3 when compared to other cpu's, but that has to do with the fact that it has been written in assembly. I am not sure whether hiarcs is entirely assembly, but considering it has mobility and such things it sure is giving the same. Also exceptions are the old 16 bits programs like kallisto, and old compiles from gnuchess. As you already can figure out, those programs are exceptions! Nearly all chess software is in c++/c and they all try to do a search and last years they all have increased in size. The number of programs that still do fit inside the L1 cache you can really see that as a limited number! In case of diep it doesn't even get close to the L2 cache when running dual :) But well, so is crafty's move generation for example. it eats memory from a total of say 900kb memory or something, not exactly something that fits inside L2 cache of modern cpu's :) In short it all comes down to the same thing. Best regards, Vincent
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.