In past postings I wrote on fast Fourier transform (FFT) performance and on GPU computing. In a new project both topics meet. We evaluated the FFT performance on GPUs. We found performance gains of more than one order of magnitude as compared to traditional (non-parallel) CPU codes.
The FFT is a core algorithm that finds applications in many different computational problems. Thus, performing FFTs on GPUs may boost many applications. In a recent paper (see arXiv:1012.3911) we showed how to solve the time-dependent Schrödinger equation and the time-dependent Dirac equations by a Fourier split operator method on GPU hardware.