Forum > GPU crunching
Latest drivers (NVidia and ATI)
sunu:
Get them from http://lunatics.kwsn.net/index.php?module=Downloads;sa=dlview;id=208
You'll also need the latest nvidia drivers.
Pappa:
--- Quote from: popandbob on 23 Jul 2009, 04:25:32 am ---On a GTX260 I just did a shorty in 1minute 20 seconds wall time... an improvement of... 2.1x faster!!
Now the question is... are they valid??
It seems almost too good to be true
-Bob
--- End quote ---
What I have running in "Seti Beta" is having "no" issues with stock or other unnoted users that were still running Optimized after the release. That is including the 2.3 DLL's introduced yesterday after it was released into Seti Main.
popandbob:
Well if the results are valid I'm seeing a HUGE improvement.
I am seeing wall times of 10.5 minutes for a long task (increased sensitivity), 5.5 minutes for an non increased sensitivity long task, and 1.3 for a short non increased sensitivity shorty and 3 min for a increased one.
Previously with 2.2 it was 8 min for a long task and 3 for a shorty.
Improvement for shorty's = 2x faster
Improvement for longer task = 1.5x faster
Bob
P.S. I estimate the final DCF at below 0.0085 since right now it takes 10 min for a task which estimates 2.5hrs @ DCF 0.17
Sutaru Tsureku:
--- Quote from: Jason G on 22 Jul 2009, 10:38:27 am ---Get only the needed 2.3 dlls in beta downloads here, and get 190.38 driver from nVidia. Some small speedup at high angle ranges is observed with V12, will be experimenting to see if new compiler makes a better build, or not, for a while.
--- End quote ---
I took the CUDA_V2.3 (.dll's) from the 'public DL area' here.
They are the same like in the 'beta release area'?
Or how I could understand your post?
You would like to optimize the '.dll's' or the 'CUDA_app'?
@ opt. crew
I have an other question because of the 'new Enhanced WUs' from SETI@home.
Could you look to this thread in the SETI@home-NC forum?
http://setiathome.berkeley.edu/forum_thread.php?id=54764
Because of new longer MB WUs..
The VLARkill option of the CUDA_V12_app will identify VLARs like before?
Or maybe.. no longer VLARkill needed, because the VLARs will be faster calculated with CUDA_V2.3 ?
Or this is only possible with a new CUDA_app?
O.K., one more question to the opt. crew.. ;)
For only CUDA crunching and best performance, PhysX ON/OFF ?
Over the system tray icon from nVIDIA.. right click -> performance and quality setting / system performance -> I can choose:
* high performance
* performance
* quality
* high quality
Stock is 'quality'.
How should be the setting for best CUDA performance?
Thanks! :)
popandbob:
VLAR's are just as slow as before... or well maybe a little slower even?
I ran one as a test in stand alone mode and well after 20 min it was only 0.35% complete.
As to my previous speed claims I've now had a couple tasks take 18 min... I tried to see how to tell the difference in the work unit header but I couldn't see anything that made sense.
I did also have an issue before when a task didn't end right the next tasks would take 20 min but I'm not sure that's the case this time because one finished in between that was a shorty at 2 minutes. Its also a possibility that there is a lot more speed variation depending on AR now than before.
Bob
Navigation
[0] Message Index
[#] Next page
[*] Previous page
Go to full version