I think the thing most nostalgic for me is that this is like 1991 all over again, except back then it was "hashes per second" climbing up from the threes and fours and tens to hundreds and thousands and eventually millions and billions.
Cc: @solardiz .
Alec Muffett
OK so I must say that LLM processing with a H200 has its merits for R&D. From one running project: "Avg prompt throughput: 6605.5 tokens/s, Avg generation throughput: 1238.8 tokens/s".
