Ollama Windows FAQ — all common questions answered

Download OllamaSetup.exe from the download page, run the installer, and open a new terminal. Verify with ollama --version. Full guide: Install on Windows.

Yes. Ollama is free and open source (MIT license). The models it runs are also free to download. There is no subscription, API key or usage limit.

Ollama supports Windows 10 and Windows 11 (64-bit). It also runs on macOS and Linux.

8 GB RAM minimum, 64-bit Windows 10 or 11, and at least 10 GB free disk space for one model. A GPU is optional but strongly recommended for usable speed. See system requirements.

Yes. The installer registers Ollama as a user-level Windows service that starts automatically on login. You can see it in the system tray. To disable auto-start, right-click the tray icon and adjust startup settings, or remove it from Task Manager > Startup apps.

Ollama supports Llama 3, Mistral, Qwen 2.5, Gemma 2, Phi-4, DeepSeek, CodeLlama and hundreds more. Browse the full list at ollama.com/library or on the Models Hub.

Models range from ~1 GB (Gemma 2 2B) to 40+ GB (Llama 3 70B). A typical 7B model takes 4–5 GB. Models are stored in C:\Users\YourName\.ollama\models.

Run ollama rm modelname. For example: ollama rm llama3. To see all installed models: ollama list. The model files are deleted immediately.

Yes. Ollama can run multiple models simultaneously if you have enough VRAM or RAM. Use ollama ps to see running models and ollama stop modelname to unload one.

Models are stored in C:\Users\YourName\.ollama\models. You can move this folder by setting the OLLAMA_MODELS environment variable to a different path. See Models location guide.

No, but response speed without a GPU is slow (5–15 tokens/s on CPU). A mid-range NVIDIA GPU with 8+ GB VRAM gives 40–80 tokens/s. AMD GPUs are supported via DirectML.

Run ollama run modelname --verbose and type a prompt. Check the gpu layers value in the stats. If it is 0, GPU is not active. See GPU Acceleration guide.

NVIDIA CUDA is generally faster than AMD DirectML for the same price bracket. CUDA has more mature optimisations in Ollama. AMD DirectML works well and is improving with each Ollama release.

Top options: (1) Enable GPU acceleration. (2) Use a smaller quantization: ollama pull llama3:8b-instruct-q4_K_M. (3) Keep models on an NVMe SSD. (4) Close GPU-heavy apps to free VRAM. See Benchmarks.

Yes. After downloading a model with ollama pull, Ollama runs with no internet connection. Prompts and responses never leave your machine. See Offline & Privacy guide.

Ollama does not send prompts, responses or personal data to any server. Inference runs entirely on your local hardware. For site analytics see our Privacy Policy.

Ollama does not persist conversation history between sessions by default. Each session starts fresh. Within a session, context is held in RAM only. No conversation logs are written to disk.

Close all terminal windows and open a new one after installing Ollama. PATH changes require a fresh session. If it still fails, see Not recognized fix guide.

Check the Windows Event Viewer for errors, or look at Ollama logs. Common causes: port 11434 already in use, antivirus blocking the process, or corrupted install. See Troubleshooting guide.

Quit Ollama from the system tray, download the new installer, and run it over the existing installation. Models are preserved. See Update guide.

Uninstall via Settings > Apps, then manually delete C:\Users\YourName\.ollama to remove models and config. See Uninstall guide.