PSO-GPU: accelerating particle swarm optimization in CUDA-based graphics processing units

This work presents a PSO implemention in CUDA architecture, aiming to speed up the algorithm on problems which has large amounts of data. PSO-GPU algorithm was designed to customization, in order to adapt for any problem that can be solved by a PSO algorithm. By implementing PSO using CUDA architecture, each processing core of the GPU will be responsible for a portion of the overall processing operation, where each one of these pieces are handled and executed in a massive parallel enviroment, opening the possibility for solving problems that require a large processing load in considerably less time. In order to evaluate the performance of PSO-GPU algorithm two functions were used, both global optimization problems, where without constraints (Griewank function) and other considering constraints, the Welded Beam Design (WBD).