Subnormal Numbers and Underflow

Updated 15 Sep 2026

•
What it is
- •
  numbers too small to be represented in the normal $(1+m) \times 2^{e-\text{bias}}$ form (with an implicit leading 1); represented instead with an implicit leading 0 once the exponent field hits its minimum value
•
Thresholds (single precision)
- •
  smallest normalized positive number: $1.0 \times 2^{-126} \approx 1.175 \times 10^{-38}$
- •
  smallest subnormal positive number: $2^{-149} \approx 1.401 \times 10^{-45}$
- •
  anything smaller than the subnormal minimum flushes to zero (underflow)
•
Related
- •
  IEEE 754 Floating-Point Standard

Linked references 2