-2

I am trying to use a base image for docker using Ubuntu 20.04 and Cuda 11.2 (as it is on my machine) to be used with tensorflow 2.5.

I modified a previous version of Ubuntu 18.04 I had. This dockerfile works fine:

FROM nvidia/cuda:10.0-runtime-ubuntu18.04
CMD nvidia-smi

but this one does not (which is a direct transcription of the previous one):

FROM nvidia/cuda:11.2-runtime-ubuntu20.04
CMD nvidia-smi

manifest for nvidia/cuda:11.2-runtime-ubuntu20.04 not found: manifest unknown: manifest unknown

neither do this:

FROM nvidia/11.2.0-runtime-ubuntu18.04
CMD nvidia-smi

pull access denied for nvidia/11.2.0-runtime-ubuntu18.04, repository does not exist or may require 'docker login': denied: requested access to the resource is denied

I concluded that this is the case here: there is no relevant image to the docker server since the message is misleading. But in the documentation is appears valid.

The same is true about this one:

FROM nvidia/11.2.0-runtime-ubuntu20.04
CMD nvidia-smi

So, can someone provide some advice over how can I use a cuda 11.2+Ubuntu 20.04 image?

Eypros
  • 5,370
  • 6
  • 42
  • 75

2 Answers2

2

Basically all of the examples except the first one are referring to non-existent images.

Here is a wrong version number:

FROM nvidia/cuda:11.2-runtime-ubuntu20.04

Here you've lost image name (cuda):

FROM nvidia/11.2.0-runtime-ubuntu18.04

Should be cuda:11.2.0-...:

FROM nvidia/cuda:11.2.0-runtime-ubuntu20.04

The list of available versions is available at Docker Hub. Also, instead of using Dockerfile, you can check if an image exists with this short command:

sudo docker run --rm --gpus all nvidia/cuda:11.2.0-runtime-ubuntu20.04 nvidia-smi
anemyte
  • 17,618
  • 1
  • 24
  • 45
1

The image you are looking for is available here.

You have missed to add the proper image name. It should be nvidia/cuda:11.2.0-runtime-ubuntu20.04. You have missed the cuda: part.

RrR-
  • 1,251
  • 3
  • 15
  • 32