Docker Image for: Python 3.6 + NVIDIA (CUDA & cudNN) drivers + TensorFlow
This is a docker image that supports python running
on a host that (optionally) has a GPU. GPU support requires nvidia-docker (2.0)
to be installed on the host system.
The primary use case of this image is to run a docker container
that supports both the CPU and GPU versions of TensorFlow. Both the CPU and GPU versions are installed in their own
virtual environments (/.cpu-env and /.gpu-env respectively), where those environments
can utilize system site packages.
The documentation for TensorFlow indicates that you must choose either CPU support
or GPU support when installing (Installing TensorFlow).
This approach provides both in one Docker image.
This is essentially a mashup of the python docker image and the nvidia-docker image.
See dockerutils for more details on docker
patterns for both CPU and GPU.