Hi everyone. It seems that if you want GPU access from a Docker container which has Linux OS, you need to have Linux OS on the host machine also. One option is to have dual-boot machine, with both Windows 10 and Ubuntu. Anyone doing something similar? I upload image for details. I want to use TensorFlow + Keras + Python in Jupyter Notebook, from a Docker container, but the GPU version of TensorFlow, that is why I ask this. Thank you.
You need to have a Linux host, but not for the reason given in this post. First reason is that Docker Desktop runs in Hyper-V and because of that is isolated from the host hardware. Second reason is that ThensorFlow uses nvidia-docker for GPU access where you can see in the prerequisites that it only works on Linux.
Me personally, I don’t like dual-boot machines, but to buy a second set of GPUs could be too expensive for such a use case.
Take a look to the Tensor Flow instruction for Docker. If you have newer version of Docker (19.3 -*) it is quit easy.