Author: Tom Kerrigan
Date: 02:10:25 05/26/03
Go up one level in this thread
On May 25, 2003 at 21:47:22, Aaron Gordon wrote: >On May 25, 2003 at 15:08:39, Matthew White wrote: > >>On May 25, 2003 at 09:23:42, Aaron Gordon wrote: >> >>>On May 24, 2003 at 16:53:57, Matthew White wrote: >>> >>>>On May 23, 2003 at 23:40:23, Aaron Gordon wrote: >>>> >>>>>>What are they measuring? >>>>>> >>>>>>IE running two copies _should_ see each copy run about 1/2 as fast with SMT >>>>>>on, since each copy is getting roughly 50% of available cpu core resources >>>>>>when running the same instruction streams. >>>>>> >>>>>>Or do you mean something else? >>>>> >>>>>It's keys per second for RC5, nodes per second for OGR. What I mean is it spawns >>>>>1 thread per processor (or virtual processor in HT's case). >>>>>My Dual Celeron 400MHz box gets an exact 2.00x speedup with this, I'm assuming >>>>>because it doesn't hit the main memory at all. Any other dual processor system >>>>>should also get a 2.00x speedup as well.. however I saw some results that were >>>>>puzzling. Here they are... >>>>> >>>>>Dual Xeon (P4) 2.0GHz without HT - 2.8 million keys/sec per thread (5.6mk/s >>>>>total), 2 threads total. >>>>> >>>>>Dual Xeon (P4) 2.0GHz with HT enabled - 0.72 million keys/sec per thread >>>>>(2.88mk/s total), 4 threads total. >>>>> >>>>>I was just wondering if you could run the same tests and confirm this. I would >>>>>have figured RC5/OGR would have managed a nice boost from HT, so it's surprising >>>>>to me. >>>>I have seen similarly poor performance on my Dual Xeon 2.4 GHz with HT enabled. >>>>I manually configured the client to only spawn 2 threads. >>>> >>>>Matt >>> >>>Did you try OGR too or did you only experience the extreme slowdown with HT >>>enabled only with RC5? If OGR was affected in any way, what was the speed >>>increase/decrease? Thanks >>I just reran the test. With 2 threads, I got 8.6M nodes/sec/thread on OGR and >>2.6M keys/sec/thread on RC5-72. With 4 threads, I got 6.6M nodes/sec/thread OGR >>and 1.1M keys/sec/thread on RC5-72. So, OGR does show increased performance with >>HT, but RC5-72 does not. I only run RC5 cracking, though, since I am much more >>interested in cryptography than the Optimal Goloumb Rulers problem :). >> >>Regards, >>Matt > >Cool, thats what I was expecting. Looks like you get a 53% increase out of OGR. >Wonder whats up with RC5... As I've been saying in the other threads, it shouldn't be that much of a surprise to see a big slowdown when running a program on one logical processor. You're sort of giving it half a Pentium 4. Prescott will be doubling most of the P4's execution resources, thereby improving HT performance. -Tom
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.