high performance computing on graphics processing units: hgpu.org

Beamforming algorithms make high demands on the computer hardware and the computation time is an important factor for the assessment of this method. This paper describes techniques for optimizing the implementation of beamforming algorithms in regard to calculation time. The main focus is on using the Graphic Processing Unit for accelerating beamforming. After a brief introduction to general purpose GPU computing, the realization of time-domain, frequency-domain and order-domain beamforming on the GPU is explained. Several benchmarks have been run and the results show that all three algorithms can gain shorter calculation time on the GPU. The tests also examine the differences in performance of the two main development toolkits, CUDA and OpenCL. Furthermore, the impact of the GPU hardware is analysed regarding computational power and scalability.