Is there a way to force pytorch to use the CPU?

Part 1 (2018)
1

Running the notebooks on a Windows 10 machine with an unsupported GPU and getting this error:

Found GPU0 GeForce GTX 960M which is of cuda capability 5.0.
    PyTorch no longer supports this GPU because it is too old.

Tried activating the fastai-cpu environment but it still tried to use the GPU (and thus gives me the error).

Is there a way to simply disable the GPU from being used by pytorch?

2

SOLVED: You need to uninstall the referenced version of pytorch in the fastai-cpu environment and install the correct version from peterjc123 repo.

conda uninstall pytorch
conda install -c peterjc123 pytorch-cpu
4 Likes
3

A quick addendum. You could use CUDA_VISIBLE_DEVICES="" jupyter notebook or CUDA_VISIBLE_DEVICES="" python main.py to have the same effect. You could replace “” with any other index like 0, 1 to refer to which all gpus the function will have access to as well.

5 Likes
4

Updated answer for fastai v1

you can force the library to use the CPU by simply setting defaults.device = 'cpu'.

WIthout wildcard imports: fastai.torch_core.defaults.device = 'gpu'

Relevant documentation: torch_core - docs

8 Likes
5

This is way easier