search

tom-doerr / TecoGAN-Docker

This is a fork of the TecoGAN project (https://github.com/thunil/TecoGAN) that adds support for docker.
Apache License 2.0
105 stars 21 forks source link

CPU support? #7

Closed ChameleonScales closed 3 years ago

ChameleonScales commented 3 years ago

In the readme, under "Install the NVIDIA Container Toolkit" it says:

This step [...] is only necessary if you want GPU support.

So if I understand correctly, we can chose to use the CPU instead? If so, I didn't find how to do it.

I tried removing --gpus all from the docker run command but then when launching runGan.py 1 I got :

ImportError: libcuda.so.1: cannot open shared object file: No such file or directory

Failed to load the native TensorFlow runtime.

If using the CPU on the run case 1 is possible, I would like to do it for the following reasons:

  1. the Nvidia Toolkit is not yet supported under the latest stable Ubuntu (20.10)
  2. I have another computer with an AMD GPU which I would also like to run TecoGAN on
tom-doerr commented 3 years ago

The issue might be that the docker base image is build using a gpu base image. In docker/Dockerfile you could change from tensorflow/tensorflow:1.8.0-gpu-py3 to from tensorflow/tensorflow:1.8.0-py3, execute the docker build step and runGan.py 1 should work on the CPU.

ChameleonScales commented 3 years ago

Thanks, it works great!