Skip to content
This repository has been archived by the owner on Apr 20, 2019. It is now read-only.

Error accessing the CUDA device

Julian Kniephoff edited this page Jun 21, 2018 · 1 revision

If you get an error to the effect that the program can not access the CUDA device/the GPU, it might help to run some CUDA program as root first. This can most simply be achieved using the following command:

sudo THEANO_FLAGS=device=gpu,floatX=float32 python3 -c "import theano"

Here, you should replace gpu with the device the error message is complaining about.

You might also look at this StackOverflow for a potential permanent solution; however, I have not tried this myself.

Clone this wiki locally