GPUs

Planned — later. No GPU package is built yet.

GPUs serve on bare metal, in containers and in full VMs. Not in microVMs.

Where a GPU works

Bare metal: the driver on the host.
Container: rootless podman with CDI, the Container Device Interface.
VM: QEMU q35 or Cloud Hypervisor with VFIO passthrough.
Not microVMs: Firecracker and QEMU microvm have no VFIO. A microVM is CPU-only.

NVIDIA

Debian 13 ships the NVIDIA 550 driver, too old for the newest cards. x0ne OS follows Pop!_OS practice:

a dedicated NVIDIA image, with the driver installed;
the open kernel modules;
a driver branch held for a whole release, changed by a human.

The opt-in x0ne-gpu-nvidia brings the driver and the NVIDIA container toolkit. The driver branch and the source repository are not decided yet.

a GPU in a container, with CDI planned

$ podman run --rm --device nvidia.com/gpu=all IMAGE nvidia-smi

Secure Boot

x0ne OS keeps Debian's signed boot chain. Modules built on your machine are signed with a Machine Owner Key (MOK), enrolled once at reboot. The policy for cloud NVIDIA images is still open.

AMD

x0ne Engine has no release yet. It will run on CPU and NVIDIA GPUs first. AMD support is planned: Debian 13 has ROCm 5.7, and Mesa from backports brings recent Vulkan drivers. ROCm workloads run in containers.

How to verify

check planned

$ nvidia-smi
$ vulkaninfo --summary

How to undo

sudo apt purge x0ne-gpu-nvidia, then reboot to unload the modules.