Author: Oliver Roese
Date: 09:05:54 02/13/02
Go up one level in this thread
On February 13, 2002 at 04:00:21, Artem Pyatakov wrote: >Hello everyone! > >I have not posted here in a while, because of all the work, but here goes. > >I am currently working on a research paper that compares the performance of >various processors combined with various compilers - in particular, I am >attempting to identify the weaker areas of each chip and compiler as related to >bit-board-based chess programs. > >I would like to use many detailed "profiles" of Crafty (gproff, MSVC performance >analyzer, NuMega, etc.) to get an idea of different areas in which these >compilers and chips over/underperform. That's where I need help - I don't really >have access to any modern computer (I just have a 450 MHz PII), and I know that >people on this site must have done profiles of Crafty (or can do them), so if >you could post some of them here, I would really appreciate that. > >To be more precise, the best profile would come from running Crafty from an >initial position to a certain depth OR solving an EPD. To emphasize, I do not >just want the kNPS for the program, I actually want a detailed profile of the >run - which functions where called how many times and how much % of total >running time did they take (or some analogue of that). The most valuable results >would be those comparing something like gcc results with results of MSVC, but >others are also very appreciated. > >I appreciate ALL the help in advance - I think that once we compare these >results, it will point to specific weaknesses of each compiler and architecture >and hopefully help us build better compilers + chips. > >Thanks. > >Artem > >P.S. If you are reluctant to post a large profile on-line please feel free to >just e-mail it to me (pyatakov@princeton.edu) Hi! Maybe you should devise a sheme how to do the profiles. Most of the folks here would be willing to help you, i think, since they do something comparable all the time long, namely matching their chess-engines each other. However most of them are not programmers and need some assistance. You could provide protocols, faqs, hints, shellscripts or such that. Some sort of plugin solution that would work with several chessengines. That would have the additional benefit to make the results more comparable. Since i use new and old memory together at home, i had to downclock my motherboard. That gets me an overall benefit but a rather large performancehit with chessengines. I learned from that, in order to make performance-prediction, one has to know the precise hardware, the clockrate of the chip AND the clockrate of the motherboard, since that is also the clockrate of your memorysystem. (Although i cant really tell about this stuff...) Under linux some informations can be read from the pseudo files under /proc. Let me say it again: Many of the folks here (not me btw) are fascinated by chessengines, but they dont really can bring them to use. Someone who devise a practical way how to do that, would be welcome. Have fun Oliver
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.