Author: Vincent Diepeveen
Date: 08:42:34 09/29/02
Go up one level in this thread
On September 28, 2002 at 03:54:14, Aaron Gordon wrote: >On September 27, 2002 at 23:42:03, Robert Hyatt wrote: > >>On September 26, 2002 at 21:04:12, Aaron Gordon wrote: >> >>>Slate ran many tests, we managed to get the speedup with one CPU using the >>>single-cpu binary vs dual cpu binary to 1.65x. Using one CPU in the dual cpu >>>binary vs two cpus in the dual cpu binary was 1.69x. In the particular tests you >>>are speaking of, was this the benchmark in crafty or some sort of test suite? >>>Also, which binary did you use for the tests? If you didn't use one of >>>mine or something very similar try going to http://speedycpu.dyndns.org/crafty/ >>>and use one of the SMP binaries near the top. I'd like to see what sort of >>>speedup you get. >> >> >>I didn't run the SMP tests for AMD, I don't have a one here and have no plans >>to get one. I posted a chart of data others provided. I don't even remember >>which position we used now. All that was significant was that all the speedup >>numbers (raw nps, not parallel search times) were in the 1.4-1.5 range with >>AMD, and 1.8 and above for the intel boxes... >> >>I personally believe it highlights a memory bottleneck... > >Perhaps, but with my binaries the AMD systems get almost 1.7x & not 1.4-1.5. 1.7 >to 1.8x speedup isn't a lot and with AMD cpus being faster I believe the dual >systems will beat any dual P4. Even some results I saw on here placed a dual >xeon 2.4 a little bit SLOWER than a single XP @ 2.52ghz. :) You must compare the same things. Of course a dual Xeon is relatively spoken faster for crafty than a dual K7. AMD is a great cpu for stand alone systems and even if you have a program that's well written. But for big bandwidth between different threads, the K7 is nothing more than a joke of course. The bottom line is that crafty is using a form of parallellism which is simply not from today anymore. It's focussing upon memory latency speeds and whatever you try, you won't make up for it when processors get faster. because the real problem in all the compares is the absolute processor speed. the xeon looks 'better' simply because it's a hell slower cpu than the AMD. Slower cpu means simply that the speed of the memory is less of a problem. obviously if you do 600k nps a second single cpu that hurts more for the memory latency than 300k nps. That's the obvious thing we forget here. Even then it's a fact that the L2 caches (or even L3 caches) from intel cpu's are performing better seemingly than from AMD. Yet AMD has way bigger L1 cache which performs great. So AMD is great for a program that doesn't poke too much into the other processor. DIEP is such a program and the majority of the dual programs are not too bad here. Crafty is the exception.
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.