All this progress has been fueled by a sustained rhythm of GPU innovation. These processing units continue to become faster and more flexible in their use. Today's GPUs can process enormous amounts of data and are used not only for rendering 3D scenes, but also for processing images or performing massively parallel computing, such as financial statistics or terrain analysis for finding new oil fields.

Whether they are used for computing or graphics, GPUs need a software interface to drive them, and we are in the midst of an important transition. The new generation of APIs brings additional orthogonality and exposes new capabilities such as generating geometry programmatically. On the computing side, the CUDA architecture lets developers use a C-like language to perform computing tasks rather than forcing the programmer to use the graphics pipeline. This architecture will allow developers without a graphics background to tap into the immense potential of the GPU.

More than 200 chapters were submitted by the GPU programming community, covering a large spectrum of GPU usage ranging from pure 3D rendering to nongraphics applications. Each of them went through a rigorous review process conducted both by NVIDIA's engineers and by external reviewers.

We were able to include 41 chapters, each of which went through another review, during which feedback from the editors and peer reviewers often significantly improved the content. Unfortunately, we could not include some excellent chapters, simply due to the space restriction of the book. It was difficult to establish the final table of contents, but we would like to thank everyone who sent a submission.