GPU's: They're Not Just for your Grandma to Play Quake, Anymore
Nick Triantos
Over the past 5 years, GPUs have grown substantially in their
programmability, precision, and performance. They've now surpassed CPUs
as the highest-performing compute engine in a computer, and the
performance difference is continuing to widen. As they've moved to
32-bit floating-point math, they've also now become practical for other
kinds of computations, from signal processing, to fluid dynamics, to
financial simulation. In this presentation, we will look at an overview
of the graphics pipeline, and see how it can now be viewed as a more
generic parallel processor of streams of floating-point data. We'll
also talk about how the programming model is likely to evolve, so that
software engineers can express the parallelism within their algorithms
more efficiently, both for GPUs, and for the upcoming wave of multi-core
CPUs.