[Feature]: Add GPU Node support

[alistarle]

Problem to Solve

To support larger use-case (including AI), I think the provider should be able to manipulate GPU instance and provide GPU worker in a Kubernetes Cluster.

Proposed Solution

Change the OscMachine controller to allow orchestration of the Flexible GPU

Additional Context

No response

Environment Details

- Kubernetes version (`kubectl version`):
- OS (`/etc/os-release`):
- Kernel (`uname -a`):
- cluster-api-provider-outscale version:
- cluster-api version:
- Install tools:
- Kubernetes distribution:
- Kubernetes distribution version:

[jdauphant-dinum]

👋 @jfbus do you think we can have an easy implementation of that ?

[jfbus]

GPU support is currently being planned, probably landing beginning of 2026. We still need to decide wether to implement it using DRA or directly in CAPOSC.

[jdauphant-dinum]

GPU support is currently being planned, probably landing beginning of 2026. We still need to decide wether to implement it using DRA or directly in CAPOSC.

Thanks for your answer.
How the DRA implementation could be done ? inside https://github.com/outscale/cloud-provider-osc ? or we can do ourself with few efforts ?

[jfbus]

How the DRA implementation could be done ? inside https://github.com/outscale/cloud-provider-osc ? or we can do ourself with few efforts ?

I'm not a big fan of adding DRA to the CCM, as both have very different scopes. A CSI sidecar could be an option, as DRA and CSI are quite similar conceptually. another option would be a completely separate deployment.

[jfbus]

Adding GPU nodes is possible thanks to #686 and the new GPU instance types. We will need to evaluate if this if the right solution or if adding fGPU support is necessary.