Solid-liquid phase change is a common physical phenomenon in our life. Moreover, it is also a hot research field, which involves the latent heat storage (used in the industrial waste heat recovery system, the building energy saving, and the solar power system), the crystallization of crystals, the solidification processing in metals and alloys, and so on. The research on the solid-liquid phase change has never been hung. Besides experiment, numerical methods are another study tools. The lattice Boltzmann (LB) method has drawn much attention in the field of solid-liquid phase change owing to its natural parallel and transient characteristic. Compared with the CPU parallel, GPU parallel is more suitable for some scientific calculations. Unfortunately, the internal memory of GPU is much less than that of CPU, which limit the number of the gird size, especially for three-dimension simulation. In order to reduce the internal memory demand, we propose a fractional-step LB method for solidliquid phase change. Compared with the general LB model, the present one needs less internal memory with the same mesh number. The analytical solutions and published numerical solutions are compared with the present results for verification. Results shows that the present model is able to simulate solid-liquid phase change.