GPU support for ARO - Batch 1

[sakthi-vetrivel]

Today, ARO does not support GPU-enabled worker nodes for an ARO cluster. This issue tracks support for them as ARO worker nodes.

NC4as T4 v3 NC8as T4 v3 NC16as T4 v3 NC464as T4 v3

[waynedovey]

Is there any moment on this? We are seeing a lot of requirements to have this feature available. Thanks

[thegovind]

Is there an ETA on when this will be in public preview?

[philipp1992]

any news on this?

[thomasphall]

Any updates?

[supernovae]

Any update on this?

[rahulm23]

We are treating this request as part of two milestones. First milestone will require more steps on part of customers to configure. At the moment it is targeted for Q1 2022. Regarding 2nd milestone, it will be a better experience and will share the ETA once we get closer.

[msftnadavbh]

Any updates?

[philipp1992]

Just use aks

[redhatstuart]

If you require this capability in ARO, please reach out to me @waynedovey , @git2g, @philipp1992, @supernovae, @msftnadavbh . Thank you.

[philipp1992]

Yes we do.I guess that's the reason people come to this issue.

[supernovae]

I wrote up a guide to implementing GPU on ARO - https://mobb.ninja/docs/aro/gpu/ - Documentation is in progress.

[jboutaud]

The instance types planned for support are the following: NC4as T4 v3 NC8as T4 v3 NC16as T4 v3 NC464as T4 v3

[jboutaud]

This is now officially supported. Here is a reference to the article explaining how to use GPUs: https://docs.microsoft.com/en-us/azure/openshift/howto-gpu-workloads

[maulik-modi22]

These T4 GPUs may not be suitable for heavy duty training workload.....

We require Tesla V100 or Nvidia A10 GPU support, eagerly awaiting for #276