How does FP8 training work on Hopper GPUs, and how do you keep it numerically stable?

FP8 promises to roughly double training throughput over BF16, but with only a handful of mantissa bits the numerics are unforgiving. Per-tensor scaling and a selective recipe are what make it converge.

Updated Aug 2026 · Grounded in real Applied AI Engineer interview loops and written to a senior-engineer editorial bar.