Author: Charles Worthington
Date: 13:55:21 02/18/03
Go up one level in this thread
On February 18, 2003 at 16:39:49, Robert Hyatt wrote: >On February 18, 2003 at 16:08:49, Charles Worthington wrote: > >>Bob, the test results from the Intel site show a hyperthreading speed-up between >>1.07 and 1.37 for various types of applications. I am no programmer but I am >>speculating that since chess engines are designed specifically to run on >>mulitiple threads and since they are known to monopolize 100% of cpu resources >>that they will be rather high on this multiplier scale. What are your >>observations with Crafty? Have you come up with a multiplier based on your own >>observations? > > >I haven't done any _exhaustive_ testing. I ran one test suite when I first got >the machine >and the raw NPS went up by 30%. I think Eugene also posted a similar number. I >then >ran a different test suite of 24 positions and the raw NPS was up by something >just over >20%. Both of these were run prior to my fiddling with the spinlock code in >crafty to >add the pause ASM instruction, although I am not sure that made a very >significant difference >after using it a bit. > >I'll try to run the tests again, the only drawback is that it is necessary to >re-boot the machine >to turn it on/off, which means I have to be in my office to do the f2=setup >stuff, which is not >so convenient. I hate to shut down during the day, and although I can reboot >from home, I >can't change the BIOS settings from home... > >It does work, however. Eugene reported that running two copies of the tablebase >compression/ >decompression code at one time speeds up by a factor of 2x. (!!) which is >_outstanding_. I have >not tried to test _anything_ other than crafty. In linux, make -j would be a >good test for seeing >how multiple compiles speed up, etc... Thank you Bob! I will stick with the 30% assumption for now but doing so will throw the assumed df7 kNs above the 3000 mark. I am very anxious to run the test to see if the figures hold out. If so, It will clobber the new generation of AMD processors. Most people here are relying on old AMD -v- Xeon figures which are from the pre-hyperthreading days and taken from machines with 400MHz FSB. I think it is time to give Intel the credit they are due and test this for ourselves instead of relying on test results from third party sources. From my own extensive observations the AMD systems have been faster in the past running single thread engines than the Intel processors. But the introduction of hyperthreading in the P4 platform may change that as well. Thanks for the info Bob!
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.