Floating-point numbers

Instead of storing all digits in a large number, store a few significant digits and a shorter exponent
$\text{significand}\times {\text{base}}^{\text{exponent}}$

• Precision - number of bits for the significand
  • single-precision fp 32 bit : 24 bits for significand
  • double-precision fp 64 bit : 53 bits for significand
• Range - number of bits for the exponent
  • single-precision fp 32 bit : 8 bits for exponent
  • double-precision fp 64 bit : 11 bits for exponent, 1 bit for sign
• Can trade off range of numerical calculation at the cost of decreasing accuracy.