Advantages of evolutionary computation with very large population for many-objective optimization problems are investigated. Effects of the population size are investigated up to 1,000,000 while the number of generations is fixed to 100. To overcome difficulty in computational time, we use a many-objective evolutionary algorithm designed for massive parallelization (CHEETAH) and use the supercomputer K. As for unimodal test problems DTLZ2 and DTLZ4, IGD property are improved up to population size 1,000,000 while GD property is saturated at population size of 10,000. Even when the total number of evaluations is fixed, this conclusion stays same. As for multimodal test problems DTLZ1 and DTLZ3, GD and IGD properties are improved up to population size 10,000 while they are not drastically improved with population size larger than that. It is probably due to the difficulty in obtaining good Pareto-optimal solutions of DTLZ1 and DTLZ3 with the current CHEETAH, which bases on NSGA-II.