Enable and validate NVIDIA GPU workloads on the media server #8

New issue

Closed

opened 2026-08-29 00:22:49 -05:00 by portegumi · 1 comment

portegumi commented

2026-08-29 00:22:49 -05:00

Collaborator

Description

Enable Kubernetes to discover and schedule the media server's RTX 3080 Ti before GPU-bound applications are considered for migration. The Sager GTX 1070 can be evaluated later if it provides useful capacity.

Scope

Select compatible host driver and Kubernetes device-plugin or operator versions.
Label and, if appropriate, taint GPU-capable nodes.
Expose GPU resources to Kubernetes without migrating ComfyUI or Jellyfin.
Record the compatibility and upgrade approach.

Acceptance criteria

The media server remains Ready with its RTX 3080 Ti advertised as a schedulable resource.
A disposable validation workload can request and use the GPU.
A workload requesting a GPU does not schedule on a non-GPU node.
Driver, runtime, and Kubernetes integration versions are documented.
GTX 1070 enablement is either validated separately or recorded as deferred.

Dependencies / blockers

Requires the media server to be joined to the cluster. Application migrations and the final UDS GPU integration are out of scope.

## Description Enable Kubernetes to discover and schedule the media server's RTX 3080 Ti before GPU-bound applications are considered for migration. The Sager GTX 1070 can be evaluated later if it provides useful capacity. ## Scope - Select compatible host driver and Kubernetes device-plugin or operator versions. - Label and, if appropriate, taint GPU-capable nodes. - Expose GPU resources to Kubernetes without migrating ComfyUI or Jellyfin. - Record the compatibility and upgrade approach. ## Acceptance criteria - [ ] The media server remains `Ready` with its RTX 3080 Ti advertised as a schedulable resource. - [ ] A disposable validation workload can request and use the GPU. - [ ] A workload requesting a GPU does not schedule on a non-GPU node. - [ ] Driver, runtime, and Kubernetes integration versions are documented. - [ ] GTX 1070 enablement is either validated separately or recorded as deferred. ## Dependencies / blockers Requires the media server to be joined to the cluster. Application migrations and the final UDS GPU integration are out of scope.

portegumi added the

gpu

cluster

labels

2026-08-29 00:22:49 -05:00