In the smooth mass distribution model, the critical curve is a line with magnification divergence on the image plane in the strong gravitational lens system. After considering the microlensing of a very small amount of discrete mass, the magnification map in the source plane has a complex structure, which provides an effective way to detect the dark matter. There are difficulties including the magnification divergence at critical curve and the huge amount of computation in microlensing simulation. To achieve the required simulation accuracy, using inverse ray-shooting directly will require huge computational resources. Therefore we developed a Graphics Processing Unit (GPU)-based parallel method to achieve the goal for massive computing. It takes about 7000 seconds to calculate calculate ～ 13000 microlenses for 10¹³ orders of magnitude of rays on the GPU of NVIDIA Tesla V100S PCIe 32 GB. On the basis of GPU parallel,compared with the direct ray tracing algorithm, the interpolation approximation improves the computing speed by about two orders of magnitude. We use this method to generate 80 magnification maps, and select 800 light curves for a statistical analysis of microcaustic density and peak magnification.