View source for Floating point processor

A '''floating point processor''' (often, technically, a '''floating point unit''') was, until recently, an option which allowed a computer to handle [[floating point]] numbers in [[hardware]]. In [[architecture|architectural]] terms, it might be a [[co-processor]], or on [[microcode]]d machines, it might be extra microcode, sometimes along with some additional hardware to speed up some aspects of the computation.

In early machines, the hardware involved was a complete extra set of [[printed circuit board]]s; later on, with the rise of [[microprocessor]]s, it became an optional chip. Now that [[Central Processing Unit|CPU]] [[integrated circuit|chips]] typically contain extremely large numbers of [[transistor]]s, floating point is usually supported in them by default.

Originally, in most CPUs with a only an integer [[Arithmetic logic unit|ALU]], integer multiply instructions took many CPU [[cycle]]s, because they were implemented by repeated shifts and adds; similarly, divides were handled by repeated subtractions. In modern CPUs where the FPU is integrated into the CPU, there is usually a multiply network which can also be used to perform these integer operations in a few CPU cycles.

Very recently, [[Graphic Processing Unit]]s have been able to handle floating point operations needed for math intensive tasks such as image rendering, etc.

[[Category: CPU Basics]]