CUDA environment path variables are used by several packages, so it makes sense to install them in the base cuda image rather than the Tensorflow one.
I found I could install gpuR on top of cuda-dev, but not cuda. OpenCl is part of the NVIDIA cuda developer libraries so it didn't require an additional install. It did require the tweak at the bottom of the cuda-dev Dockerfile, which I found here: https://github.com/cdeterman/gpuR/issues/52.
CUDA environment path variables are used by several packages, so it makes sense to install them in the base cuda image rather than the Tensorflow one.
I found I could install gpuR on top of
cuda-dev
, but notcuda
. OpenCl is part of the NVIDIA cuda developer libraries so it didn't require an additional install. It did require the tweak at the bottom of the cuda-dev Dockerfile, which I found here: https://github.com/cdeterman/gpuR/issues/52.This includes my other PR.