high performance computing on graphics processing units: hgpu.org

With the evolution of graphics processing units (GPUs) in floating point operations and programmability, GPU has increasingly become powerful and cost-efficient computing architectures, its range of application has expanded tremendously, especially in the area of computational simulation. In this article, the Fourier method combined with GPU acceleration techniques is applied to simulate large-scale transient temperature field in cooling control. Although it is possible to perform temperature field simulation on a personal computer through Fourier method, when grids are huge, a tremendous CPU calculating time is required which is unacceptable. Thus GPU accelerating technique is used for the parallel processing of Fourier method and a significant speedup can be observed. Following the programming model of compute unified device architecture (CUDA), the iteration process of Fourier method is improved into several kernel functions by the single instruction multiple thread (SIMT) mode and multiple processors of the GPU execute these kernel functions. Numerical results with over 13 speedups demonstrate the efficiency of GPU computing technique of the Fourier method. The absolute error between GPU and CPU is less than 10-12 in double-precision.