Solving sparse matrices: A comparative analysis between FPGA and GPU

Abstract

In this paper, the performance improvement from using FPGAs and GPUs to solve sparse matrices is evaluated. GPU and FPGA are formally compatible but the results are very sensitive to the particular calculation task, the architecture of DSP blocks in particular FPGA, and the architecture of the designed core for particular calculation in FPGA. The results show that GPUs have superior performance over FPGAs for this particular type of problems.