Geek Answers Handbook

How to diagnose a failure to start CUDA due to lack of resources?

I get an error outside the resources when I try to start the CUDA kernel (via PyCUDA), and I'm wondering if it is possible for the system to tell me which resource I have short on. Obviously, the system knows which resource has been exhausted, I just want to request this.

I used an employment calculator, and everything looks good, so either where the corner was not closed, or I'm wrong. I know that these are not registers (which seems like a common criminal), because I use <= 63, and it still does not work with the 1x1x1 block and 1x1 grid on the CC 2.1 device.

Thanks for any help. I placed the thread on the NVidia boards:

http://forums.nvidia.com/index.php?showtopic=206261&st=0

But there were no answers. If the answer is "you cannot query the system for this information," that would be good to know (sort of ...;).

Edit:

The most popular use of the register that I saw was 63. Edited above to reflect this.

+5
source share
2 answers

, PyCUDA API- CUDA, - : CUDA_ERROR_LAUNCH_OUT_OF_RESOURCES , cuLaunch() . PyCUDA, , , , , , .

, ...

+6

.

CUDA : sm_12 vs sm_20

, 70 .

0

More articles:


All Articles