"We introduced an effective algorithm for k-nearest neighbor problem which works on multiple GPUs. By an experiment, we have shown that it runs more than 330 times faster than an implementation on a single core of an up-to-date CPU. We have also shown that the algorithm is effective from the viewpoint of parallelism of GPUs. That is because 1) there is no synchronization between GPUs until the very end of the process and 2) the workload is well balanced."

"Huang at Nvidia has said many times that CUDA is one of the most important advances in computing, and its promise will be evident as programmers learn how to make use of it. Nvidia is already getting lots of its chips designed into supercomputers and servers, thanks to CUDA. And because of CUDA, the 480 chip can do physics processing 2.5 times faster than the previous generation. That means that the environment in a game, such as water in a stream, behaves far more realistically, adding to the overall illusion of a graphics animation. The real question is whether CUDA is really helping or hurting Nvidia’s cause to bring better graphics to the entire world."