Install Llama.cpp on Windows, pick the right GPU build (CUDA or Vulkan), and run your first LLM with llama-server.