[throwawaymaths]

maybe but it shouldn't be surprising. cerebras's designs were born ~2014 ~pre transformers, and the megachips were initially targetted for hpc workloads. it was definitely "solution looking for a problem" back then and now is drifting into square peg in round hole territory now (see sibling comment about groq). I'm surprised they have gotten their raw perf as high as they have by now.

[arisAlexis]

Ilya was very much in awe and totally contrary to what you are saying btw if you read the Elon ope ai emails. Also, they do run llama the fastest.

[throwawaymaths]

in the real world fastest isnt everything. power and capital cost per token is