question

AndreasChristofi-8362 avatar image
0 Votes"
AndreasChristofi-8362 asked ·

Docker image building for GPU workload on AKS cluster

Hello, I've had some trouble building an image that can utilize the GPU on the AKS cluster. The instructions shown on this article were followed to install the Nvidia plugin. My understanding is that the image needs to contain the Nvidia drivers, therefore I've used the nvidia/cuda as the base image and tried to install specifically python 3.7.6 on it. However, the final image always ends up with python version 3.6.9.

Also, which cuda version should I install?

I would appreciate any suggestions on how that can be done properly.

Moreover, I'd like to know how to specify the Nvidia library locations in the Kubernetes manifest, as mentioned in this article.

Thanks!

azure-kubernetes-service
· 2
10 |1000 characters needed characters left characters exceeded

Up to 10 attachments (including images) can be used with a maximum of 3.0 MiB each and 30.0 MiB total.

@ AndreasChristofi-8362 Apologies for the delay in response and all the inconvenience caused because of the issue. I have reached out to our internal team on this and will keep you posted once I have an update.Thank you for your patience over the matter.

1 Vote 1 ·

Update: I've managed to specify the python version in the image.

As I mentioned in the original message, the instructions were followed to install the Nvidia device plugin on the cluster.
I'd like to know the required steps for enabling the docker image to access the GPU.

0 Votes 0 ·

1 Answer

AndreasChristofi-8362 avatar image
0 Votes"
AndreasChristofi-8362 answered ·

Update:

I've managed to make it work by using this Nvidia base image. Thereafter, installed python dependencies and updated the default python interpreter to be the version that I need. Also, the python-cffi package needs to be installed. Hope this helps anyone trying to do the same!


·
10 |1000 characters needed characters left characters exceeded

Up to 10 attachments (including images) can be used with a maximum of 3.0 MiB each and 30.0 MiB total.