PDA

View Full Version : post lunatics / cuda install issues.


none1
03-30-10, 05:45 PM
HI!

I've a machine I've been running the unoptimized vanilla berkley setiathome client on, cpu client only, with no issues, quad core intel, dual geforce 280GTX-M processors.

I followed the sticky directions. Downloaded the lunatics installer, downloaded the CUDA dll's, thought I was good to go.

1) I now get some workunits that report out as "computation errors" were previously I did not get these.

2) I only see FOUR tasks active, just like prior to enabling CUDA, but I expected to see 6 (quad cpu + dual GPU). All 4 tasks say application "setiathome_enhanced 6.03". How do I tell if CUDA is working?

3) I have browsed around, and seen some folks jumping through hoops to get multiple GPU's to work. Is there any tips or a guide to make this happen?

Any help appreciated, I want to make sure I'm getting the most WU out of this machine as possible.

Thanks!

muddocktor
03-30-10, 07:18 PM
Have you updated to the latest drivers? If they don't have the CUDA apps in the driver package, it will error out like that. I was setting up CUDA on a watercooled Q6600 system that I just got back up and running while the cpu is doing Rosetta and I forgot to update to the latest drivers instead of the generic W7 drivers and had this problem. After installing the latest drivers, I went ahead and reinstalled the CUDA optimisations and all is well.

You might have problems though, since you want to run CUDA on a laptop. I haven't ever tried that myself. You might check nVidia's website and see what kind of mobile drivers they have for your gpu.

Kryten451
03-30-10, 07:30 PM
As I understand it, it is by design. The computation errors are only for WUs that have been rejected by Lunatics CUDA drivers "varkill" feature which allows only optimal types of WUs for CUDA processing. The WUs showing errors will disappear from the list and your CUDA processing will not be burdened by non-optimal WUs.

donuts
03-30-10, 07:50 PM
I'm currently out of CUDA units. I ran my cache out to test drive folding, now I can't get more.
Getting SETI to run on a CUDA card was fairly painless, after following the guide in the sticky, it recognized my GTX 295 and my gx2 and launched 2 per card. (these are on separate machines)

Unless " use GPU while computer is in use" is ticked in the BOINC manager, you won't see them, just the started units.

Also check your "messages" tab in the BOINC manager, currently I'm getting "project has no jobs available" I only run SETI on my gpu's, Rosetta on the cpu's

none1
03-30-10, 11:13 PM
OK I think donuts got part of it right ... I now see FIVE processes, 4 for seti and ONE for seti (cuda). Ya looking at the logs it looks like the first few hours after I activated cuda there were no jobs available /facepalm

BUT, it does now look like only ONE of my GPUs is running, anyone know how to get multiple GPU's running under CUDA ?

THANKS!