Docker doesn’t use GPU

  cuda, docker, linux, python, tensorflow

I’m trying to write a usable Dockerfile for linux GPU prediction with python (I’m new in docker).

Config:

  • os (Ubuntu or other linux distribution)
  • cuda (9-10) (I have gtx 1080 ti)
  • python 3.6
  • tensorflow-gpu==1.9

I tried to use nvidia/cuda images from docker hub like: 9.2-cudnn7-devel-ubuntu18.04

It runs, with –gpus=all parameter, but the python code doesn’t utilize the GPU.
What could be the problem?

Source: Docker Questions

LEAVE A COMMENT