3

Like many othersm I'm getting a Runtime error of Cuda out of memory, but for some reason pytorch has reserved a large amount of it.

RuntimeError: CUDA out of memory. Tried to allocate 2.00 MiB (GPU 0; 6.00 GiB total capacity; 4.31 GiB already allocated; 844.80 KiB free; 4.71 GiB reserved in total by PyTorch)

I've tried the torch.cuda.empy_cache(), but this isn't working either and none of the other CUDA out of memory posts have helped me either.

When I've checked my gpu usage(nvidia-smi) before running my python program, it is plenty free.

Improve this question
1

2 Answers 2

Reset to default
2

From the given description it seems that the problem is not allocated memory by Pytorch so far before the execution but cuda ran out of memory while allocating the data that means the 4.31GB got already allocated (not cached) but failed to allocate the 2MB last block. Possible solution already worked for me, is to decrease the batch size, hope that helps!

Improve this answer
Sign up to request clarification or add additional context in comments.

Comments

1

You should find processes by typing,

!ps -aux|grep python

and then when you find the process you should kill it by typing,

!kill -9 652

This will kill process number 652. In your case this will be something else that you want to get rid of.

NOTE: Remember you have to start over your code if you kill some process that you should not have ended. But this is the most easiest and manual way to do it.

Another Note: Or you can always decrease the batch size, if the problem happens again after successfully emptying the gpu cache.

Improve this answer

Comments

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.