Native Arch Linux ARM and Omarchy for NVIDIA DGX Spark, with ports of its NVIDIA compute and development tools.
Developer preview: v0.1.0-preview.3. This is a source release for developers, not an installer or a supported replacement for DGX OS. One physical Spark runs it from the internal NVMe beside DGX OS, with accelerated Omarchy, native CUDA, GPU containers and an AI Workbench GPU project. Full factory software parity is still in progress.
| Component | Verified behavior |
|---|---|
| Boot and storage | Native ARM64 kernel, external USB SSD root, Ethernet and SSH |
| Omarchy | Login, Hyprland/Quickshell, NVIDIA acceleration at 3840×2160 |
| CUDA | Native compilation and kernel execution; 4 MiB unified-memory verification |
| Containers | Docker with NVIDIA runtime; CUDA execution inside containers |
| AI Workbench | Native backend, ARM64 project build, GPU execution, JupyterLab with token enforcement verified (authenticated API; no-token and wrong-token refused); repeated after backend restart |
| DGX Dashboard | Web service, firmware inventory and read-only Arch package status |
| Networking and audio | Four RDMA devices, Bluetooth controller and HDMI audio sink detected |
See hardware validation for versions, test scope and limitations. Container workloads run on the native Arch host; Workbench itself is not hidden inside an Ubuntu VM.
- Getting started
- Build and test the source
- Known issues and remaining work
- Dual boot on the internal NVMe
- Which Install menu entries work on ARM
- Release notes
- Publication review and post-publication reassessment
- Upstream provenance and licenses
- Artwork: hero, announcement card, wallpaper and icon
- Contributing
- Requirements for a supported release
packages/ contains Arch recipes and compatibility helpers; patches/ contains
upstream changes; manifests/ and upstream-lock.json record source locations and
pins. Tests include host-side regression checks and explicitly separate native
integration probes; passing them does not certify the operating system, the
vendor software or any installed machine. image/make-disk-image.py assembles
regular files only; it is not an installer and does not configure a bootable
root filesystem for you.
The preview deliberately ships no boot image, vendor binary packages, firmware, private inventories, credentials, or automated raw-disk writer. NVIDIA payloads are downloaded from their upstream locations when you build the recipes, under their own terms. Original project code is MIT licensed; third-party material retains its applicable licenses.
This is an independent community project, not an official NVIDIA or Omarchy release. Keep a working recovery path. Kernel, graphics, firmware and rolling Arch updates require testing as a set; an ordinary package upgrade is not yet a validated update/rollback workflow for this port.