WikiBeck

Passing a GPU through to a TrueNAS Scale app

3 min read

Passing a GPU through to a TrueNAS Scale app

TrueNAS Scale can hand a system’s GPU to an individual app (e.g. Jellyfin or Plex for hardware transcoding) without full PCI passthrough or GPU isolation — those mechanisms are for a different scenario (dedicating the GPU exclusively to a VM). This guide covers the simple app-level GPU selection and a workaround for a TrueNAS Scale 24.10 bug that could break it.

Prerequisites

  • TrueNAS Scale 22.12 or later.
  • A GPU suitable for the workload (NVIDIA or AMD) already installed and, if the host is itself a VM, passed through to it at the hypervisor level (e.g. Proxmox PCI passthrough) with a working driver bound inside the guest.
  • The app to be configured already installed.

Steps

1. Select the GPU in the app’s settings

TrueNAS Scale ships with the necessary GPU drivers preinstalled. Open the target app’s Resources/GPU Configuration settings and select the desired GPU directly.

Do not enable general PCI passthrough or GPU isolation for this — select only the GPU in the app’s own resource settings. Those other options are for handing the GPU to a VM instead, and are unrelated to (and conflict with) simple per-app GPU use.

2. Workaround for the TrueNAS Scale 24.10 GPU-selection bug

TrueNAS Scale 24.10 has a known bug where the GPU cannot be selected for an app through the UI at all, failing with:

Expected [uuid] to be set for GPU in slot [0000:02:00.0] in

Work around it via the midclt CLI instead of the UI.

Identify the available GPU(s) and their UUIDs/PCI slots:

midclt call app.gpu_choices | jq

Apply the GPU assignment directly to the app’s config:

midclt call -j app.update <app-name> '{"values": {"resources": {"gpus": {"use_all_gpus": false, "nvidia_gpu_selection": {"<pci-slot>": {"use_gpu": true, "uuid": "<gpu-uuid>"}}}}}}'

For example (illustrative values — substitute the app name, PCI slot and UUID from the previous step’s output):

midclt call -j app.update jellyfin '{"values": {"resources": {"gpus": {"use_all_gpus": false, "nvidia_gpu_selection": {"0000:02:00.0": {"use_gpu": true, "uuid": "GPU-7cbffdba-cee8-ab11-5d68-987dd76c57e1"}}}}}}'

Verify

Reopen the app’s resource settings in the UI — the GPU should now show as assigned. Check the app’s own hardware-acceleration status (e.g. Jellyfin’s Dashboard → Playback, or nvidia-smi inside the app’s container if shell access is available) to confirm it’s actually being used.

Troubleshooting

  • GPU selection fails with Expected [uuid] to be set for GPU in slot ...: this is the known 24.10 bug — use the midclt workaround above instead of the UI picker.
  • GPU doesn’t appear in app.gpu_choices at all: this points to a driver/passthrough problem one layer below app-level selection — check that the GPU’s driver is actually bound inside TrueNAS itself (e.g. nvidia-smi at the host shell) before troubleshooting the app config. A GPU architecture unsupported by the currently bundled driver (for example, Pascal-generation NVIDIA cards under a version that only ships the open-source kernel module) will not show up here no matter what app-level settings are tried — see NVIDIA open kernel module dropping Pascal GPU support.

Sources

Graph View

Backlinks