Back to self-hosting
Operating System

Windows Self-Hosting

Run local LLMs on Windows with WSL2, native CUDA, and tools like LM Studio and Ollama.

WindowsWSL2CUDAOllamaself-hosting

Recommended approach

  • WSL2 with Ubuntu for the cleanest Python/PyTorch/CUDA experience
  • Native Windows for GUI tools like LM Studio, KoboldCPP, or local Stable Diffusion suites

Minimal configuration

  • GPU: NVIDIA RTX 3060 12 GB
  • RAM: 32 GB
  • Storage: 512 GB NVMe SSD
  • OS: Windows 11 23H2 with WSL2 Ubuntu 24.04

Recommended configuration

  • GPU: NVIDIA RTX 4090 24 GB
  • RAM: 64 GB
  • Storage: 2 TB NVMe SSD
  • OS: Windows 11 23H2
  • Stack: WSL2 + Ollama or LM Studio

Notes

AMD ROCm on Windows is limited; NVIDIA is currently the most reliable path. Some Apple Silicon users prefer macOS for native MLX performance.