Author: Robert Hyatt
Date: 14:07:29 04/10/03
Go up one level in this thread
On April 10, 2003 at 14:51:11, Anthony Cozzie wrote: >>But I don't buy the 50% stuff, the cpu is not that simple internally. One >>thread will run at >>nearly full speed and the other gets slipped into the gaps, which is what I see >>(at least) when >>running Crafty. > >This statement seems counterintuitive to me. If one thread is given a higher >priority in the CPU, than the OS would need to take this into account (and not >just with PAUSE). Could you add a little debugging code to crafty and measure >how many nodes are computed by each thread on a run of your dual xeon? I'm very >curious if it is a 40/40/10/10 split or a 25/25/25/25 split (the latter seems >more logical to me). > >anthony I can't do it, because the threads are not bound to a logical processor, and they bounce around. I could spit out nodes searched by a single thread each time it does a single parallel search, but that would produce several thousand "spits" that would be difficult to interpret... I think the main issue is that no thread runs at 100%. Otherwise SMT would not work anyway. In the normal case one runs at well below peak theoretical performance, and the other gets slipped into the cracks, which are pretty big. But conflicts reduce this to a pretty small overall boost...
This page took 0.03 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.