CUDA and System Updates

CUDA (Compute Unified Device Architecture) is NVIDIA's parallel computing platform and programming model that gives software direct access to GPU cores. DGX Spark ships with a specific CUDA toolkit version — keeping it current ensures compatibility with the latest model-serving frameworks like vLLM and PyTorch, and unlocks performance improvements tied to newer driver releases.

Method 1 — DGX Dashboard (Recommended)

1. Find dgx-spark-dashboard.desktop on the desktop
2. Right-click → Allow Launching (first time only)
3. Double-click to open, or navigate to http://localhost
4. Log in → click "Update Now"
5. Allow system to reboot, then check for updates again — repeat until no more updates appear

Method 2 — Terminal

Run on both Sparks (directly or via SSH for Spark 2):

sudo apt update
sudo apt dist-upgrade -y
sudo fwupdmgr refresh
sudo fwupdmgr upgrade
sudo reboot

Expected Versions After Update

# Verify
nvidia-smi --query-gpu=driver_version --format=csv,noheader
# Expected: 580.159.03

nvcc --version
# Expected: release 13.0