high performance computing on graphics processing units: hgpu.org

This work presents a parallelization method for the Particle Swarm Optimization algorithm using a low-cost architecture: a General Purpose Graphics Processing Unit (GPGPU). The strategies to better suit the architecture main characteristics are addressed along success rates and convergence times for the optimization of Rastrigin’s and Ackley’s functions on a 30-dimensional search space, and compared with results previously obtained using a cluster implementation.