Designing a GPU Based Heat Conduction Algorithm by Leveraging CUDA

One-dimensional heat conduction algorithm plays a significant role in physical material research. It is normally implemented by using CPU serial programming or MPI parallel programming. Unfortunately, real applications normally process large volumes of particles, the query performance is very low when using the above two approaches. This paper designs a parallel heat conduction algorithm by leveraging the powerful parallel processing capabilities GPU. The algorithm parallelly executes the calculation of the displacement and velocity of each particle on GPU, and then sends the results to CPU, the CPU finally calculate the average heat flux and the temperature of each particle. The proposed algorithm was evaluated on the CUDA parallel programming environment. Our experimental results demonstrate that the performance of the proposed GPU based parallel algorithm significantly outperforms the traditional CPU serial implementation. We achieve a performance speed-up of 900. In terms of time reduction Furthermore, the performance speed-up grows with the increase of the particle number.