View Single Post
  #74 (permalink)  
Old 01-January-2006, 10:03 PM
HenrikOlsen's Avatar
HenrikOlsen HenrikOlsen is offline
Order of Kilopi
 
Join Date: Sep 2003
Location: Denmark 55.6773° N 12.3610° E
Posts: 9,764
Send a message via MSN to HenrikOlsen Send a message via Yahoo to HenrikOlsen
Default

Quote:
Originally Posted by Ken Vogt
Henrik: Welcome to the thread!

I'm sure you are correct in this. I believe the folks at CPDN said that running 2 WUs simultaneously on an Intel HT is about 15% faster than running two consecutively.
Thanks.
I should probably have mentioned that I'm basing my statement on a series of experiments on HT performance I ran a while ago when I was involved with debugging and tuning a fairly large cluster.
The result showed that the performance gain from running two processes on one HT cpu varied a lot with the software run, with everything from <5% to >50% depending on how well the software was optimised for the cpu.
The better the software was optimised, the less the gain from running two copies.
15% gain sounds like it's quite optimised to begin with.
__________________
‘To those who regard “crime fiction” as some sacred icon which must follow a rigid formula, I will always be the man who writes 18-syllable haiku.’
Andrew Vachss, Autobiographical essay
Trying to make sense of computers, The Error Log.
Reply With Quote