Author: Bo Persson
Date: 08:59:12 10/30/00
Go up one level in this thread
On October 30, 2000 at 07:03:18, Dan Newman wrote: >On October 29, 2000 at 09:15:28, David Rasmussen wrote: > >>When I compile my program on Windows 2000 in VC++ I get ~150000 nps on a >>Cel300A@464, but when I use the exact same compiler (or even the same >>executable) on win98 I get ~60000 nps !!! >> >>And I know its not win98 thats the problem. I have run it on win98 lots of time >>and compiled it with both gnu and borland c++ builder, at got fine results. >> >>Whats wrong here? >> >>P.S. the machine with win98 is an athlon 650 Mhz. Can it be the athlon that >>doesnt like the heavy bitboard usage ? > >I benchmarked my bitboard program, Shrike, on a number of different machines, >including an Athlon 800. Here's some of what I got: > >Cel/400 472 knps >P3/600 (Katmai) 555 knps >P3/800 (CuMine) 909 knps >Ath/800 698 knps > >This was all with the same executable, compiled with MSVC. I've wondered >what's going on with the Athlon results. One of my theories is that the >Athlon doesn't have a very fast implementation of the BSF/BSR instructions, >but I haven't looked into at all--just idle speculation. Well, it is more than just idle speculation: BSF and BSR are among the "... VectorPath instructions, which should be avoided in the ADM Athlon processor". (AMD Optimization Guide) > The Athlon (by >its specs) should be *really* good. My program really seems to like large >fast caches, so I expected it to do a lot better on the Athlon... > >If you aren't using BSF/BSR for finding first/last bit, then it must be >something else. Maybe MSVC optimizations are very Pentium specific. > >-Dan. Bo Persson bop@malmo.mail.telia.com
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.