Books on parallel programming theory often talk about such weird beasts like the PRAM model, a hypothetical hardware that would provide the programmer with a number of processors that is proportional to the input size of the problem at hand. Modern general purpose computers afford only a few processing units; four is currently a reasonable number. This limitation makes the development of highly parallel applications quite difficult to the average computer user. However, the low cost and the increasing programmability of graphics processing units, popularly know as GPUs, is contributing to over...