GPU implementation using CUDA

Abstract

This thesis considers the application of desktop computer video card as a processor to solve two algorithms in medical imaging and sparse matrix operations. The GPU (Graphic Processing Unit) hardware structure in the video card is designed and dedicated to 3D graphic rendering that include matrix and vector operation. To reconstruct the Magnetic Resonance Images, we apply IFFT that is a fast algorithm for Fourier transforms and has a parallel structure that can be used in GPU processor. Another experiment for GPU application is sparse matrix operations. Two case studies to work with sparse matrix operations are 662_bus and 494_bus admittance matrices. We apply these two matrices to obtain lines current. We Implement the algorithms on GPU GeForce GTX 295 in CUDA platform at Visual C++ Host compiler, the results show 7X speedup when the same kernels running on CPU Phentom™ II X4 2.6GHz.