Like many other optimization algorithms, particle swarm optimization could be possibly stuck in a poor region of the search space or diverge to unstable situations. For relieving such problems, this paper proposes a hybrid intelligent computing: a multi- subpopulation particle swarm optimization. It combines the coarse-grained model of evolutionary algorithms with particle swarm optimization. This study utilizes two performance measurements: the correctness and the number of iterations required for finding the optimal solution. The results are obtained by testing the particle swarm optimization and multi-subpopulation particle swarm optimization on the same set of function optimizations. According to both types of performance measurement, the multi-subpopulation particle swarm optimization shows distinctly superior performance over the particle swarm optimization does. An additional set of experiments is performed on only the hard functions by adapting the algorithm parameters. With such adaptation, the improvement succeeds. All experiments are executed without taking parallel hardware into account.