Nengo fails to load if environment is set up to use GPU but device can't be found
Closed this issue · 1 comments
In my case this occurs if I try to remote desktop to my work computer, which creates its own environment without access to the GPU or something like that. If you then try to run Nengo, you get the error message
no CUDA-capable device is detected
and then the whole process terminates. Probably what we would like to have happen is when a CUDA device can't be found, it smoothly loads in the normal fashion rather than just terminating.
Xuan and I traced it down to the hasGPU() call in WeightedCostApproximator, which we believe hooks into the checkCudaError() function in customCudaUtils.c. Thought it would be best if we leave the fix up to Eric though, since we don't really know what's going on in there.
yeah that shouldn't happen, on it.