Something that would be extremely helpful is a good benchmark of various hardware for llm inference. It's really hard to tell how well a GPU will perform or whether it will be supported at all.