Why Isn't Ollama Using My GPU?

Check this before anything else

Confirm the GPU is visible at the OS level, outside of Ollama entirely:

nvidia-smi

If that command isn't found, or shows no GPU, this isn't an Ollama problem yet - it's a driver problem. See the driver section of Building a Full PC for Local AI before going any further here.

If nvidia-smi works, check what Ollama actually sees

With a model loaded, run:

ollama ps

Look at the PROCESSOR column. If it says 100% CPU, Ollama is not using the GPU for that model, even though the GPU itself is fine. If it shows a GPU percentage (or a split like 60%/40% CPU/GPU), it's already partially or fully using it - the model may just be too large to fit entirely in VRAM, which is a different, related problem covered below.

If the model should fit, but it's still on CPU

• Restart the Ollama service after any driver update. Ollama detects the GPU at startup - if drivers were installed or updated after Ollama was already running, it may still be using the state from before. On Linux: sudo systemctl restart ollama. On Mac, quit and reopen the app from the menu bar.
• Running inside WSL2 on Windows? GPU passthrough into WSL2 has its own driver requirements separate from the regular Windows NVIDIA driver - confirm nvidia-smi works from inside the WSL2 shell specifically, not just from Windows itself.
• Check for a second, unused GPU. If the machine has both an integrated GPU and a discrete NVIDIA card, confirm Ollama isn't defaulting to the integrated one - nvidia-smi only reports on NVIDIA hardware, so it won't warn you about this.

Still stuck? The Ollama Out of Memory entry covers what happens when a model does try to use the GPU but doesn't fit.