CULA error: Runtime error

[Maja]

Hi,

I am working on a Windows XP 32-bit with two Quadro FX 580. I have CUDA 2.3 installed and Jacket v1.2. The NVIDIA driver is version 190.38.

My problem is when I run svd. I get the following error:

"Jacket:53(jktFunction) CULA error: Runtime error, see culaGetErrorInfo for error code
??? Error using ==> svd
CULA failure (save your work and restart Matlab) (svd.cu:53)"

And then I am informed where in my Matlab script the error occurs.

Any suggestions?

/Maja

[kyle]

Hi Maja,

For Jacket issues, you should deal directly with Accelereyes on their support forums: http://www.accelereyes.com/forums/

They have another post that might be able to help you as well: http://www.accelereyes.com/forums/viewt ... f=7&t=1112

My quick guess is that you are running out of memory, though. I think Jacket reports CULA out-of-memory error as less descriptive runtime error. Since you are generating two output matrices (U and V') in addition to the input matrix, you need a lot of GPU memory. If your input matrix + U + V + other GPU allocations are over the memory size of your FX580, you'll see this error. Try do some quick tests with smaller matrices.

Hope this helps or the Jacket team can assist you more.

-Kyle

[Maja]

Hi Kyle,

I think you are absolutely right. I do however have another question: Do you know if it is possible to use both GPU's simultaneously? I am very new in this field...

Thank you for your quick response!

/Maja

[kyle]

CULA doesn't currently parallelize across multiple GPUs. However, our libraries ensure thread safety allowing you to performing two CULA operations simultaneously on two different GPUs. We have an example in CULA 1.1 that demonstrates this functionality.