nvidia-docker not working with non-cuda containers

Created on 4 Jul 2018 · 1Comment · Source: NVIDIA/nvidia-docker

hi! I installed nvidia-docker on my ubuntu 18.04 machine using docker from the docker website (docker-cc). When I try a nvidia docker image all things work:

$  docker run --runtime=nvidia nvidia/cuda nvidia-smi 
Wed Jul  4 14:50:12 2018       
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 390.48                 Driver Version: 390.48                    |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|===============================+======================+======================|
|   0  GeForce GTX 1080    Off  | 00000000:08:00.0  On |                  N/A |
| 50%   58C    P0    51W / 210W |   1254MiB /  8112MiB |      1%      Default |
+-------------------------------+----------------------+----------------------+

+-----------------------------------------------------------------------------+
| Processes:                                                       GPU Memory |
|  GPU       PID   Type   Process name                             Usage      |
|=============================================================================|
+-----------------------------------------------------------------------------+

But when I try to use the standard cuda bundled with Ubuntu inside the docker image things seem to be breaking. For example this docker file:

FROM ubuntu:18.04                                                                                                       
RUN apt-get update && apt-get install -y nvidia-utils-390

gives me problems.

$  docker build . -t test                            
Sending build context to Docker daemon  2.048kB
Step 1/2 : FROM ubuntu:18.04
 ---> 113a43faa138
Step 2/2 : RUN apt-get update && apt-get install -y nvidia-utils-390
 ---> Using cache
 ---> f68df36e2de4
Successfully built f68df36e2de4
Successfully tagged test:latest
$  docker run --runtime=nvidia test nvidia-smi        
Failed to initialize NVML: Unknown Error

Also installing the full nvidia-cuda-toolkit package inside the container doesn't help. Am i missing something? can I only use the Docker files supplied by Nvidia?

Source