I have a budget of $4000. Should I get a mac studio m3 ultra or should i build my own server/desktop for LLM inference?