Receptor-ligand molecular docking aims to predict possible drug candidates for many diseases, and it requires a lot of computing cost. Shortening this time- consumption process will help pharmaceutical scientist to speed up drug development. In this paper, a parallel molecular docking simulation based on CPU-GPU heterogeneous computing is proposed. This simulation is developed from our previous developed molecular docking code iFitDock (Induced fit docking program) which introduced Non-dominated Sorting Genetic Algorithm II (NSGA II) and Molecular Mechanical-Generalized Born Surface Area (MM-GBSA) binding free energy. In this program, the most computationally intensive part is the computing of scoring functions due to complex computing process of free binding free energy. Thus, this paper focuses on offloading the computing of scoring functions as well as related conformation spatial transformation to GPU, and keeping the rest of the simulation on CPU. A detailed CPU-GPU heterogeneous computing model is constructed to parallelize the computing of scoring functions and related workload on the GPU and to define the data exchange between GPU and CPU. The primary parallel iFitDock system with only parallel semi-flexible docking implemented achieves a speedup of around ~20× with respect to a single CPU core. The result shows that it is very productive to use CPU-GPU heterogeneous computing for semi-flexible molecule docking cases in iFitDock.