IEEE floating point formats on the Intel architecture

"Format" is the word used to describe the layout of the the various parts of a floating point number (in memory or floating point register).

The meaning (semantics) of the various combinations of bits is termed the encoding.

The formats for the binary floating point numbers supported by the FPUs of the Intel 80x86 microprocessors are: single precision (4 bytes), double precision (8 bytes), and extended precision (10 bytes).

To convert a number x to a normalised number in one of these formats, write: 

      x = (-1)^S * m * 2^(E-B)
Where:
1.0 < = m < 2.0, i.e. m = 1.bbbbbbbbbbb,
S is 0 or 1,
B is the exponent bias of the particular format,
and E is an integer > = 0.

Then select a format and put: 

      [sign] = S;
      [exponent] = E;
      [significand] = bits of m, excluding leading '1' if
                      single or double format.
For example, consider the decimal number 0.1. This number has S = 0, E-B = -4, and (in binary) m = 1.1001100110011001100110... For a single precision number, this gives: 
      [sign] = 0;
      [exponent] = -4 + 127 = 123 (decimal), 01111011 (binary);
      [significand] = 10011001100110011001101
Where we have rounded the signficand to the nearest representation in 24 bits. The bytes in memory for this number would therefore be (Intel puts the least signficant byte at the lower memory address): 
      0xcd 0xcc 0xcc 0x3d

Single Precision

Known in "C" as "float". Size is 4 bytes (32 bits). Significand precision of normalised numbers is 24 bits (the most significant bit is an implied '1'). 
      [sign] [exponent] [significand]
       1 bit   8 bits      23 bits
      Exponent bias is 127.

Double Precision

Known in "C" as "double". Size is 8 bytes (64 bits). Significand precision of normalised numbers is 53 bits (the most significant bit is an implied '1'). 
      [sign] [exponent] [significand]
       1 bit   11 bits     52 bits
      Exponent bias is 1023.

Extended Precision

Known in "C" as "long double". Size is 10 bytes (80 bits). Significand precision of normalised numbers is 64 bits. 
      [sign] [exponent] [significand]
       1 bit   15 bits     64 bits
      Exponent bias is 16383.
Note that unlike the other formats, the '1' to the left of the binary point is included in the significand, i.e. a normalised extended precision number always has a '1' in the most significant bit of the significand.