Abstract: Liquid sloshing is a common phenomenon induced in partially filled tanks under external excitations, which may destroy the tank structure and vessel stability. Moving particle semi-implicit (MPS) method is a typical meshfree method which can effectively simulate violent liquid sloshing problem. However, the low computational efficiency of MPS is the bottleneck of its application in large-scale three-dimensional problems. In the past years, GPU parallel acceleration technique has been widely used in the field of scientific computing. In this work, GPU parallel acceleration technique is introduced into MPS method and an in-house solver MPSGPU-SJTU is developed by using CUDA language. Then this solver is used to simulate 3-D liquid sloshing in liquefied natural gas (LNG) tank. The convergent validation of particle spacing is carried out to verify the accuracy of present solver. The maximum particle number of simulation model is over two million particles. MPSGPU-SJTU solver can accurately predict the impact pressures by comparing with other results. In addition, the violent flow phenomena such as large deformation and nonlinear fragmentation of free surface can be observed in these simulations. The comparison of computation time between GPU and CPU solvers demonstrates GPU parallel acceleration technique can significantly reduce the computation time and improve the computational efficiency of MPS. The phenomena of liquid sloshing in LNG tank and rectangular tank are compared. The results show that LNG tank can reduce the sloshing amplitude and impact pressure in high filling level. However, the sloshing is more violent and the free surface presents three-dimensional feature in LNG tank with middle and low filling level. Finally, the investigation of the effect of different fluids such as water and LNG on sloshing phenomena is also conducted in this paper. It shows that the flow fields of both liquids are almost similar and the impact pressure is proportional to the liquid density.