Comment by pcthrowaway

One thing about the GPUs that I think is unmentioned, is that the LLM models are also becoming more efficient, right? So you can get more tokens or more "work" out of the same GPU as the technology improves