0

I’ve heard that gpus can only compute simple instructions but do them in a parallel matter, which makes them suited for machine learning.

What happens if a pytorch tensor is in gpu but the type of computation I want to perform is not included in the gpu’s instruction set? Does the data in vram travel to the cpu for that specific computation?

Danny Han
  • 177
  • 3
  • 9
  • what type of operation do you have in mind? most "complex" operations can be executed by a series of simpler ones – Shai Jan 17 '22 at 06:19
  • thank you for the answer! I was thinking, any type of complex operation that would not be sufficient with a GPU? I was just wondering how packages such as pytorch run.. – Danny Han Jan 17 '22 at 07:23
  • I think I'm struggling with a similar issue to what you're describing and I'm getting the error "RuntimeError: Unable to find a valid cuDNN algorithm to run convolution". Might be wrong though. – yann ziselman Jan 17 '22 at 09:08
  • Please provide enough code so others can better understand or reproduce the problem. – Community Jan 27 '22 at 14:47

0 Answers0