‹ BackHN Continuity

Thread

Subnormal floating-point numbers are expensive on Intel processors

69 points · 54 comments · zdw

  1. pixelpoet · · focus · HN ↗
    This has been the case since a zillion years, since the Core 2 Duo days at minimum.
    1. account42 · · focus · HN ↗
      The interesting part is that this seems to be Intel-specific.
      1. augment_me · · focus · HN ↗
        This is true on most hardware. Running ftz on H100s or B200 gives a free 10-20% boost for GEMM-epilogue workload
        1. adrian_b · · focus · HN ↗
          This is true only on most hardware that does not care about computational errors, i.e. which is intended mostly for games or for AI.
Open on Hacker News to reply ↗

Unofficial Hacker News client; not affiliated with Y Combinator.