DGX OS is the operating system that ships on every NVIDIA DGX Spark, and it is not some bespoke NVIDIA invention you have to learn from scratch. It is a lightly customised build of Ubuntu 24.04 LTS with NVIDIA's entire AI software stack already installed and configured. Power the box on and CUDA, the drivers and the main inference frameworks are simply there, ready to use.

Quick Answer

DGX OS is a Linux distribution based on Ubuntu 24.04 LTS, preloaded on the DGX Spark with CUDA 13.0, cuDNN, NCCL, TensorRT-LLM, vLLM and the NeMo framework. It exists so developers can run and serve AI models the moment they unbox the machine, without spending a day installing drivers and CUDA toolkits.

What DGX OS actually includes

The base is standard Ubuntu, so the file system, package manager and command line behave the way any Linux developer expects. What NVIDIA adds on top is the part that saves time. The GPU drivers and CUDA toolkit are matched to the GB10 superchip and preinstalled, which removes the single most common source of setup pain on a fresh AI machine.

On top of CUDA sit the libraries and frameworks: cuDNN and NCCL for the low-level maths and communication, then the model-serving engines. TensorRT-LLM compiles models into optimised engines with INT4 and FP8 support, vLLM handles high-throughput serving of many concurrent requests, and the NeMo framework covers fine-tuning. PyTorch and JupyterLab are included too, so you can open a notebook and start experimenting straight away. If you are weighing up which local-AI hardware to buy, the AI PC range at Evetech shows how these machines are configured for South African buyers.

What DGX OS does not do

This is not a general desktop operating system, and treating it like one misses the point. It is tuned for AI development and inference on DGX hardware, so it will not be your daily-driver machine for browsing, office work or gaming. The whole design goal is that the same tools you would use on a larger DGX Station or in DGX Cloud are present here, so code moves between them without rewrites.

It also will not turn a non-NVIDIA box into a DGX. DGX OS is built around the GB10 superchip and its unified memory, which is why it ships only on the Spark rather than as a download for any PC. For a more flexible machine that runs Windows and still handles capable local AI work, Evetech's best-selling GPUs are the route most local developers take.

Frequently Asked Questions

Is DGX OS just Ubuntu?

At its core, yes. It is built on Ubuntu 24.04 LTS, so the command line and package management are familiar. NVIDIA's customisation is the preinstalled AI stack and the GPU drivers tuned for the DGX Spark hardware.

Can I install DGX OS on my own PC?

No. It ships preloaded on the DGX Spark and is built around the GB10 superchip and its unified memory. It is not offered as a general download for other machines.

Which AI frameworks come preinstalled?

CUDA, cuDNN and NCCL at the low level, plus TensorRT-LLM, vLLM and the NeMo framework for serving and fine-tuning. PyTorch and JupyterLab are also included so you can start experimenting immediately.

Do I still need to install drivers myself?

No. The GPU drivers and CUDA toolkit are preinstalled and matched to the hardware, which removes the most common setup headache on a fresh AI machine.

Curious whether a local AI machine fits your work? Browse the AI PC range at Evetech to compare options built for running and serving models on your own desk.