Author: Robert Hyatt
Date: 20:20:25 09/04/02
Go up one level in this thread
On September 04, 2002 at 23:13:48, Aaron Gordon wrote: >On September 04, 2002 at 22:26:13, Robert Hyatt wrote: > >>On September 04, 2002 at 22:03:24, martin fierz wrote: >> >>>...or in the short version, 1.69 more nps for the 2-processor box. >> >> >>Totals so far: >> >>dual xeon 1700mhz 1.84X raw NPS >>quad xeon 700mhz 1.90X raw NPS (using only two processors) >>dual AMD 1730mhz 1.67X raw NPS > >Hyatt, please shed some light on this for me. Here's something I've been >noticing.. It appears the faster the chips involved the slower the speedup gets. >It seems the older machines get a 1.8-1.9x speedup where as some of the newer >machines (fast dual AMD's, some of the higher end dual P4's) get lower speedups. >Is this due to memory bandwidth limitations perhaps? What are your >thoughts on this? > >If anyone here has an older dual-amd box I'd appreciate you trying to run my >CraftyK7SMP version 18.15 binary and checking it out.. Right now I only have >an SSE version of the binary so anyone with a dual Tbird or pre-morgan Duron >won't be able to run it (if you have such a system email me and I'll compile a >binary asap). Sean Mintz, if you read the I'd appreciate it if you could run it >for me on your dual 1.2MP's and post the results. > >You can grab the binary at http://speedycpu.dyndns.org/crafty/ In general, it is a memory conflict/bandwidth problem. Quads eliminate some of this with 4-way interleaved memory. Duals don't go to that more expensive extreme and as a result, memory is a bottleneck. And it gets worse as processor speeds increase. however, note that we are only talking 10% for the tests so far, from my 190% NPS increase to the 180% increase by SW.... That's not horrible. I'm going to try to run the test on our dual PIV/2000mhz XP machine tomorrow...
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.