We present our benchmark results for a newly developed multi-phase chemodynamic code by using the raceSPH library on different hardware accelerators. A maximum speedup of 24× is reached for the SPH part in the serial run when the Smoothed Particle Hydrodynamics (SPH) calculation is accelerated by an NVIDIA GeForce 8800 GTS 512 graphics card (128 cores, 1.8GHz). This code is developed for the simulation of physical processes in dwarf galaxies, in which the SPH method is used to describe the behavior of warm gas. The raceSPH library implements the SPH calculation on different hardware. It provides a set of general interfaces to help integrate hardware accelerators into existing astrophysical programs. In this paper we show that after both gravity and SPH calculation are accelerated by GPU, the neighbour list construction consumes most of the time and becomes the bottleneck of the current program.