You laugh, but the requirements are actually very different between computer vision and other pre-2023 "AI" tasks and LLMs. LLMs are hopelessly memory bandwidt

You laugh, but the requirements are actually very different between computer vision and other pre-2023 "AI" tasks and LLMs. LLMs are hopelessly memory bandwidth bound. Most other models are compute bound.