The simple path: you probably don't need WSL2
Ollama ships a native Windows installer that talks to your NVIDIA GPU directly through CUDA, with no Linux subsystem involved. If you just want to run ollama run commands in a terminal and chat with a model, downloading the installer from Ollama's official site and running it is the entire setup. WSL2 and Docker only come into play for more advanced cases: containerized deployments, some AMD GPU compatibility workarounds, or specific development workflows. If a guide's first step is "open WSL2," you've likely landed on content aimed at a more advanced setup than you need. See the beginner guide for the same install-to-first-chat steps that work here.
The install-order mistake
Ollama checks which GPU libraries are available at the moment it's installed. If you install Ollama before your NVIDIA drivers are fully set up, it won't know your GPU exists, and it'll quietly fall back to CPU, no error message. This is the single most common cause of "Ollama isn't using my GPU" on Windows.
The fix: make sure your NVIDIA driver is current first (check with nvidia-smi in a terminal, which should print your GPU's name and driver version), then install or reinstall Ollama after that's confirmed working.
Confirming your GPU is actually being used
After installing, run a model and check with:
If it shows the model running on CPU rather than GPU, that's the install-order issue above, or your GPU doesn't have enough VRAM for the model and Ollama is offloading part of it. Check the model's real memory requirement against your GPU in the hardware calculator before assuming it's a driver problem.
A note on AMD GPUs
NVIDIA has the smoothest path on Windows. AMD GPU support through ROCm is meaningfully less mature on native Windows; if you're on AMD and running into persistent issues, a Linux install tends to be more reliable for local AI work than continuing to fight Windows-specific AMD driver quirks. See the hardware guide's GPU tier table for more on this tradeoff.