View Single Post
  #30 (permalink)  
Old September 5, 2008, 07:39 PM
LCB001's Avatar
LCB001 LCB001 is offline
Folding Captain
 
Join Date: Feb 2008
Location: Aylmer QC.
Posts: 1,774

My System Specs

Default

Quote:
Originally Posted by SugarJ View Post
The GPU client 6.20 works on both Nvidia and ATI GPU's. But my 8800GT got about 5000ppd using 10% CPU while I'm getting 1800-2400 ppd on my 4870 while using 65% CPU. I got more out of the SMP client. I'm about to give up on F@H again, concentrate on Boincing.
Don't quit folding just yet, the Ati clients are starting to get new larger WU's which make them less CPU dependent. The PPD's are starting to get up around 3850 PPD for a 4870. You might find Folding Forum • View topic - New larger WU (4742-4747 1254 atoms) interesting.

Nvidia folders should check this quote


Re: New larger WU (4742-4747 1254 atoms)

by texaslabrat on Thu Aug 28, 2008 5:15 am
The PPD difference comes from the huge differences in the dependency on the cpu between the two. If you had a liquid-nitrogen-cooled 8Ghz cpu, you could likely match/exceed the nvidia cards' PPD's with a 4870 on the smaller WU's. As the ati cores become less and less cpu-bound through mhouston's work, you'll see that the new WU's being issued for Nvidia will produce smaller and smaller PPD relative to the science being done since the benchmark machine will be doing more ns/day of simulation when the initial PPD mark is set with the more efficient core available at the time. At the end of the day, I believe we'll wind up seeing the PPD numbers matching what our expectations would be given the raw TFLOPS available and the observed efficiency rates of shader use. The Nvidia crowd should soak in the points while they can, because once the larger proteins start being crunched by them, their PPD will likely decrease drastically from what they are accustomed to now just by the fact that the ati client has become more efficient and is the benchmark reference by which everything else is compared to.
At the point where a given WU can be run on the benchmark machine in a non-cpu-bound condition is the point where a true apples-to-apples comparison between the two camps' gpu hardware (and stream computing software stack) can be made.
__________________
Folding For Team 54196

Reply With Quote