A new, extremely fast sorting algorithm has been created for GPUs (CUDA-capable only) by researchers at the University of Virginia. This new algorithm is capable of sorting at a rate of one billion (integer) keys per second using a GPU. Normally CPUs aren’t as efficient as CPUs for these types of algorithms (for sorting), but these researchers claim to have created one much faster than any other known CPU-based sorting method.

This algorithm is for sorting large sequences of fixed-length keys/values. The current implementation can sort any of the C/C++ built-in numeric types or any structured payload types. Head over to the project’s website using the link below or download the code directly and try it for yourself.

RadixSorting - back40computing (Google Code)