Intel Arc also works surprisingly fine and consistently for ML if you use llama.cpp for LLMs or Automatic for stable diffusion, it’s definitely much closer to Nvidia in terms of usability than it is to AMD