KWSN Orbiting Fortress

Mildew · Posted: Thu Oct 11, 2012 11:27 am Post subject: Cuda errors

I'm getting a lot of errors on my newest shrubber.
Anyone else have the same problem or have any idea what might be the cause?

I'm running a mighty GT 640 with 2 GB RAM.
Driver version is currently 306.81, but it failed with the previous version of the driver as well (306.23).
The application is SETI@Home Enhanced 6.10 (cuda_fermi).
Application exits with code 0x1 after 11 seconds of computation on all work units. My guess would be 11 seconds of CPU work before the GPU kicks in and the WU fails...

#ni-1

Pooh Bear 27 · Posted: Thu Oct 11, 2012 12:35 pm Post subject:

Have you tried it on any other projects?
Fermi is a style of card. I do not think the 640 is a Fermi. NVidia says it's a Kepler technology card. I think you should be running CUDA23.
_________________
Watch the movie about me The 4-1-4s: The Original Teenage Hackers (It is only about 12 minutes long.)
My lucky numbers are 121*2^4553899-1 and 3756801695685*2^666669±1

Mildew · Posted: Thu Oct 11, 2012 1:46 pm Post subject:

It's running GPU grid, Primegrid, Poem and MilkyWay without issues.

There are several versions of the 640, but I think all non-OEM versions are GK107, so a Kepler card right?

Anyway, it's not me that decide which application to use, it's done by the BOINC client. How can I force it to use the CUDA23 application? Some kind of manual download?

Pooh Bear 27 · Posted: Thu Oct 11, 2012 5:07 pm Post subject:

Try it on the Astropulse which is OpenCL on SETI and see if that works. If it does then write the admins to fix seeing the 640 card as a Fermi.

Putting_things_on_top · Posted: Fri Oct 12, 2012 10:13 pm Post subject:

The Kepler cards (most of the 600 series) are fairly new on the 'scene'.
F@H has not yet published its FAH_Core16.exe (at least for public use) yet.
Maybe other projects have not fully addressed the Fermi vs Kepler differences.

Have you searched other project-specific message boards & forums for similar complaints (and possible solutions)?

My experience with a GTX-690 (on F@H) has been an ordeal.
Just this past week, my average PPD dipped because I had some "un-cooperative" work-units that downloaded but couldn't execute.
So my 690 sat there for 3+ days doing nothing!

_________________
Click here for...KWSN F@H team summary at EOC

Or here for...KWSN F@H team overtake at EOC

Mildew · Posted: Sat Oct 13, 2012 2:16 am Post subject:

I put a question on the SETI forum, and apparently all you need to do is add an environment variable named CUDA_GRID_SIZE_COMPAT and set it to 1.

No idea what this does or why it's working, but I'm up and running again! Cool

AdrianGak · Grunt Joined: 17 May 2016 Posts: 1 Location: Slovenia

Ive started gettings WU errors too. At first it was just on my 8800GT G92 and I thought they just stopped supporting the older CUDA, but now Im starting to get them on my GTX275. I know they are concentrating on getting Fermi working right now, so maybe this is just a growing pain. Ill shut down GPU projects during WCG week to make sure no unstable apps affect it and save the 0.31 CPU usage to feed the GPU.