Bluefin GDX is the recommended ISO for this use case. It is a dedicated AI/ML workstation image built specifically for NVIDIA hardware with CUDA support.

Key features included:

• NVIDIA open drivers + full CUDA stack pre-installed
• uv (ultrafast Python package manager) pre-installed
• nvtop for GPU monitoring
• NVIDIA Container Toolkit for GPU-accelerated containers (PyTorch, TensorFlow, etc.)
• Docker, Podman, and VS Code with Dev Containers — ready to use
• RamaLama for local LLM inference with CUDA acceleration

Download: bluefin-gdx-lts-x86_64.iso

Important: GDX is a fresh-install image — do not rebase to it from an existing Bluefin or Fedora system. For hybrid GPU laptops (Intel iGPU + NVIDIA dGPU) like the Yoga Pro 9i, GDX handles the setup properly by pre-loading i915 in the initramfs for hardware acceleration on the integrated GPU.

Once installed, GPU-accelerated containers can be run directly, for example:

podman run --gpus all nvcr.io/nvidia/pytorch:26.01-py3