GoKawiilBenchmark shows subnormal floating-point math slows Intel Xeon chips far more than AMD or Arm rivals
A developer benchmarked how Intel, AMD, Arm and Apple processors handle IEEE 754 subnormal floating-point numbers—the tiny values near zero that require special handling. Testing double-precision arithmetic on Intel Granite Rapids and Emerald Rapids Xeons, AMD Zen 5, AWS Graviton, and Apple M4 Max chips, the results showed Intel processors suffer dramatic slowdowns, sometimes 40-plus times slower, when computations involve subnormal values compared to normal ones.
GoKawiil's interpretation of the reporting above, not reported fact.
Subnormal numbers commonly appear in machine learning, audio processing, and physics simulations where values gradually shrink toward zero, so this performance cliff can silently degrade real workloads on Intel hardware. Developers who assume uniform floating-point performance across CPU vendors may need to flush subnormals to zero or restructure code specifically to avoid this penalty on Intel systems.
- Intel Xeon chips (Granite Rapids and Emerald Rapids) show severe slowdowns—up to 40x—when processing subnormal floating-point numbers.
- AMD, Arm-based Graviton, and Apple M4 Max processors handle subnormal numbers with far less performance penalty in the same tests.
- Even a small fraction (1%) of subnormal values in an array can meaningfully slow down operations on affected hardware, impacting fields like ML and gaming.
Source: lemire.me — Https, 2026-09-15
Published there as: “Subnormal floating-point numbers are expensive on Intel processors”
Read the original report → The summary and analysis above are GoKawiil's own, written from reporting by the source above. Facts and quotes belong to the original publisher.