Floating Point Unit
From Unofficial BOINC Wiki
[edit] General
That portion of the computer that does the operations on Floating Point Numbers.
With integer math you have a fundamental limitation of what types of number ranges that you can represent. With 8 bits you can represent about 256 numbers (0 to 255, or -128 => 0 => 127); with 16 bits, 65,000 numbers, as in 0 to 65,536; larger numbers can be represented by extending the total number of bits that are used. The place where this breaks down is when we have a significant difference in the magnitude of the numbers that we are going to be working with.
In the old days before pocket calculators we used slide rules and systems based on logarithms to handle this very class of problems. In the computer world we created a numbering system that leveraged the number of bits we have and we have a number and exponent contained within the one binary value. Normally the smallest of these numbers are contained within a 32-bit binary number and can cover a fairly wide range with reasonable accuracy. Since the decimal point appears to move we started calling this "floating point" (whew! I bet you thought I would never get to the point).
Now, we have a number system that can represent a broad range of numbers and everything is perfect. Right? Well, almost. Floating point numbers give us a good dynamic range and a reasonable representation of numbers. However, they cannot represent all numbers with an exact value. There are "gaps" in between values and the system will choose one of the two closest values that it can use to represent the number.
The portion of the Central Processing Unit (CPU) that performs the manipulation of these values and performs the mathematical operations of these values is called the Floating Point Unit, or FPU for short. Like the integer unit, it does mathematical operations using hardware systems to give us reasonable speed in producing answers.
