high performance computing on graphics processing units: hgpu.org

Digital Particle Image Velocimetry (PIV) is an optical technique used to measure the velocity of seeded particles in real flow. A CCD camera captures the flow field twice under exposure to a short duration laser flash. Recorded image pairs are cross-correlated to extract velocity information from these records. Time resolved PIV technology can capture images with some hundreds of frames per second. In this paper, we present a PIV-system that implements vector field reconstruction and visualization on programmable graphics processing units (GPUs) thus providing a high-speed back-end for time resolved PIV technology. We propose an efficient FFT implementation on such hardware, which is used to cross-correlate multiple pairs of interrogation windows. To visualize extracted vector fields we employ functionality to create and to render geometry data on the GPU. In this way, not only can any data transfer between the CPU and the GPU be avoided, but spatial information derived from PIV as well as the time history of points in the flow can be combined instantaneously.