RTFM · NVIDIA Open
NVIDIA Open modules and the planned N-Spawn of Saphira
nvidia-open 610.57.04 is staged in the hatchling repository for a pure-musl Saphira host. The host-module milestone is proven without a GPU; real RTX validation and CUDA-in-nspawn remain pending.
The host boundary: pure musl
Saphira's host remains a pure musl system. It carries nvidia-open 610.57.04 kernel modules and the matching GSP firmware, but it does not install CUDA, glibc, or NVIDIA proprietary userspace. This keeps the host boundary explicit: kernel modules and firmware on the host; any NVIDIA userspace outside it.
| Place | Contents | Status |
|---|---|---|
| Saphira host | musl; nvidia-open 610.57.04 kernel modules; matching GSP firmware | Host module path staged in hatchling |
| glibc systemd-nspawn container | CUDA and NVIDIA proprietary userspace matching the host nvidia-open release exactly | Architecture/design only; not tested |
Do not add glibc compatibility or CUDA to Saphira itself. The planned container boundary is not a reason to change the host libc policy.
What Hatchling has proved
Milestone 1 was proven on Hatchling, which has no NVIDIA GPU: the nvidia-open modules load, NVRM initialises, PCI probing correctly reports no NVIDIA device, and the driver unwinds cleanly. That proves this no-device host path only; it is not evidence of a working GPU, CUDA workload, or container setup.
# /etc/apk/repositories
https://packages.akadata.ltd/saphira/hatchling
# Refresh Hatchling, then confirm the staged package is visible.
apk update
apk list | grep nvidia
# nvidia-open-610.57.04-r0 x86_64 {nvidia-open} (MIT OR GPL-2.0) [installed]Prove it works: Correct interpretation of this milestone
A clean no-GPU load, initialisation, probe, and unwind is recorded. Real hardware support is not claimed until the RTX 3060 / GA106 test is complete.
Next hardware proof: Homer
The prepared USB-safe test on Homer is the next milestone. It is intended to validate the real RTX 3060 / GA106 path against the staged host package. Until that test is completed and recorded, the real-GPU result remains pending.
This documentation deliberately records the distinction between a no-device kernel-module test and a real GPU result. It does not infer one from the other.
Planned N-Spawn of Saphira
The proposed CUDA path is a glibc systemd-nspawn container: an N-Spawn of Saphira. CUDA and NVIDIA proprietary userspace belong in that container, never on the Saphira host. The container's NVIDIA userspace must match the host nvidia-open release exactly; for the currently staged host package, that means 610.57.04.
GPU device nodes under /dev/nvidia* must be passed into the container as required by the NVIDIA userspace. The exact container image, device-node arrangement, and CUDA validation have not yet been tested. Treat this as the established architecture and test plan, not as an operating recipe or a working CUDA claim.
- Keep the Saphira host musl-only: no glibc, CUDA, or NVIDIA proprietary userspace.
- Use NVIDIA userspace in the glibc nspawn container that exactly matches the host nvidia-open release.
- Pass the required /dev/nvidia* device nodes into the container.
- Record a separate real-GPU and CUDA-in-nspawn proof before calling the container route supported.
Kernel-package policy
Saphira's current policy is rebuilt, versioned kernel-module packages when kernels bump, following the same package model used for ZFS. DKMS is not part of this plan unless a concrete reason emerges.
Did we miss something?
If this page left something unanswered, found an error, or there is another subject you would like documented, tell us. Saphira’s documentation grows from real problems people need to solve.
Send feedback or request a new section →
Prefer not to do it yourself?
Everything needed to do the work yourself is documented here and remains free — we charge for human time, not for withholding knowledge. Sometimes the missing resource is simply time. The same people who build Saphira can provide paid professional help with implementation, migration, troubleshooting and administration.