← Field notes
Field note / 01Kernels
What we turned off to make Helios faster on one H100
The fastest-looking option on paper did not survive the second minute of a run.
FP8 looked like the headline win until minute two of every run, when accumulated quantization error collapsed the stream. It ships disabled.
What held instead: FlashAttention-3, a compiled hot path, and a playout ladder that stopped throwing generated frames away.
Then there is the box lottery. Nominally identical H100s do not behave alike, so an engine-verify gate now runs before any box takes traffic.
Windflow EngineeringRuntime team
Field notes are short on purpose: one decision, what it cost, what we kept. The long write-ups land here when they are ready.