System for representing Floating Point values; kind of scientific notation with extra steps

What it looks like

For a 32 bit floating point.

3231 - 2322 - 0
0000000000000000000000000000000
SignExponentMantissa

Special Cases

E == 00 < E < 255E == 255
M == 00Powers of 2
M != 0non-normalized, typically overflowOrdinary numbersNot a number (NaN)