The LINPACK Benchmark on a Multi-Core Multi-FPGA System
Emanuel Ramalho · TSpace (University of Toronto) · 2008
The LINPACK Benchmark is used to rank the most powerful computers in the world. This thesis is an implementation of the benchmark on a multi-FPGA system to see how the performance compares to the processor-based implementation. TMD-MPI is the MPI implementation used to parallelize the software portion of the algorithm while the TMD-MPE provides the same functionality for the hardware engines. Results show that, when using small sets of data, one FPGA can provide a speedup of 1.94 over a high-end processor running the LINPACK Benchmark with Level 1 BLAS. However, there is still opportunity to do better, especially when scaling to larger systems. ii Dedication I would like to dedicate this thesis to all of the people that have made it possible. To my parents for all of their love, support and financial effort throughout the past two years and without whom this would have never been possible. To my girlfriend Jas, whom I love from the bottom of my heart, that has supported me in the good and the not so good moments and has made this last year seem like a dream. To all my family, but especially to my grandma, “Xica ” that has always encouraged me to strive forward and that passed away during my stay here in Canada. Grandma “Xica”, you will always be in a special place in my heart. I would also like to thank my supervisor Professor Paul Chow for sharing his knowledge with me, for his patience, guidance and feedback that helped me achieve my objectives. Special thanks to all my research group for all of their help, comments and availability whenever