[Official FAQ] Corrections and follow-up questions welcome.

NVIDIA CUDA remains the path of least resistance. ROCm, Arc, and laptop NPUs improve, but drivers and operator coverage vary.

Buy/build advice:

• Search for your exact GPU + backend + model success reports before purchasing.
• NPUs often suit smaller models/assistive acceleration, not desktop-flagship expectations.
• Vulkan/Metal backends can be the pragmatic way to “just run” on non-CUDA hardware.

Please post model, quant, driver, and tokens/s.