In this paper, the characteristics of the typical two models of parallel genetic algorithms are compared. Those models are the coarse grained model and the micro grained model. Especially, the parallel efficiency and the total calculation time on PC clusters that are built with commodity hardware are examined. The characteristics are examined through the numerical examples. There are two major characteristics in the coarse grained model. One of them as the network cost is very small. The other is the fact that the necessary number of iteration is smaller than that of the model of the micro grained model. On the other hand, in the micro grained model, the ideal parallel efficiency cannot be reached to 100%. Therefore, it is concluded that the coarse grained model is suitable for PC cluster systems.