NVIDIA DGX Spark Setup Guide: From Unboxing to Your First Local Model

Step 1: first boot

Plug in the supplied power adapter and Ethernet. You can set up with a monitor and keyboard, or headless — the unit can broadcast a temporary Wi-Fi hotspot for browser-based setup (details are on the card in the box). Create your user account and connect to your network.

Step 2: update everything

Install DGX OS, firmware and driver updates before installing tools, then reboot.

• sudo apt update && sudo apt full-upgrade
• Reboot, then confirm the GPU with nvidia-smi.

Step 3: remote access

Most people use DGX Spark from another computer. NVIDIA Sync connects your laptop to the Spark and launches tools like VS Code, terminals and web dashboards; plain SSH works too. Give the unit a fixed IP on your router to make this reliable.

Step 4: run your first model

Pick one runtime to start.

Step 5: add an agent stack

Point your agent framework or coding assistant at the local OpenAI-compatible endpoint (Ollama on port 11434, vLLM on 8000). Start with one agent, check memory with nvidia-smi, then add more.

Common setup problems

Most issues have quick fixes.

• Can't find the unit on the network → check the router's device list and assign a fixed IP.
• Model won't load → it is too large; choose a smaller quantization.
• Slow first response → the model is loading from disk; later replies are faster.
• Container can't see the GPU → install updates and use NVIDIA's container toolkit.