A parallel algorithm is proposed for modeling of crowd multi-cell occupancy and velocity variety. Methods are devised to solve issues such as space competitions so that each pedestrian can autonomously determines his or her own movement. As pedestrians are modeled independently and simultaneously, the proposed algorithm can be easily implemented into heterogeneous computing frameworks such as OpenCL, CUDA etc, which support large scale parallel computation, and preserve these frameworks’ high performance while the issues are efficiently solved. In the section of numerical experiments, the proposed algorithm is firstly validated through two different experiments. Then a quite thorough investigation of impacts of multi-cell occupancy and velocity variety upon crowd macroscopic dynamics is applied.