[Official FAQ] Share your experience or follow-up questions below.

All three work well. The right choice depends on your goals.

• Native Windows: Ollama and LM Studio have installers and work well once your NVIDIA driver is installed. A good choice for personal use and learning.
• Windows with WSL2: useful if you need Linux-first tools such as vLLM or training frameworks. Key point: install the NVIDIA driver on Windows only. Do not install a separate Linux GPU driver inside WSL, since WSL uses the host driver.
• Linux: the usual choice for servers and production, with the most complete driver, CUDA and container ecosystem.

Checklist

• Run nvidia-smi. If it shows your GPU and driver version, the driver is working.
• Use a virtual environment (venv or conda) for Python tools to avoid dependency conflicts.
• Keep PyTorch, CUDA and driver versions compatible. Use the install command from the official PyTorch page.
• For Docker on Linux, install the NVIDIA Container Toolkit and start containers with --gpus all.
• Reserve plenty of disk space. Single models can be several GB to tens of GB, and you can usually point the tool to a larger drive.

Common problem: if everything runs on CPU even though you have a GPU, the usual causes are a driver or CUDA mismatch, or installing a CPU-only build of a package.

Which OS are you on and where are you stuck? Paste the command and its output.