Art. 08 – Vol. 21 – No. 2 – 2011
Universitatea “Politehnica” Bucureşti
Abstract: Graphic processors (GPUs), originally designed exclusively for game industry and visual effects, have turned into complex multi-threaded architectures which are now closer than ever to gain the co-processor statute. GPUs can be used to process data and to deal with computationally intensive applications really fast. Programmers’ access to these platforms has been enhanced along with the official launch of the GPGPU dedicated architectures. This paper focuses on the study of the performances offered by one of the most competitive architecture, CUDA, used to elaborate optimized image processing algorithms. This case-study presents the results gathered running a parallel algorithm designed to compute a 256 gray level image histogram on CUDA platform and it points out the performance differences between the results collected when the algorithm was run on the CPU vs. on the GPU platform.
Key words: Graphical processing unit (GPU), CUDA, parallel computing, GPGPU.