Subnormal floating-point numbers are expensive on Intel processors
Intel CPUs incur significant performance penalties when processing subnormal floating-point values, which are numbers very close to zero. The article explains that handling these values requires extra micro‑code steps, slowing execution compared to normal numbers. It discusses the impact on scientific and graphics workloads, and notes that software can avoid the slowdown by flushing subnormals to zero or using alternative algorithms.