Coalesced computations of the incompressible Navier–Stokes equations over an airfoil using graphics processing units

This paper presents a Graphics Processing Unit (GPU) based implementation of the Finite Differencing Time Domain (FDTD) methods, for solving unsteady incompressible viscous flow over an airfoil using the Stream function-Vorticity formulation for a structured grid. For the large-scale simulations, FDTD methods can be computationally expensive and require considerable amount of time to solve on traditional CPUs. On the contrary, modern GPGPUs such GTX 480 are designed to accelerate lots of independent calculations due to advantage of their highly parallel architecture. In present work, the main purpose is to show a new configuration for leveraging GPU processing power for the computationally expensive simulations based on explicit FDTD method and CUDA language. Our proposed work improves the GPU FDTD results by increasing the global memory coalescence with the same amount of occupancy, resulting in an increase in maximum output performance. In addition, this study introduces a more coalesced pattern of data loading which reduces the global memory requests. Although both GPU based programs are over 28 times faster than a sequential CPU based version, Implementation of our proposed work showed up to 44% decrease in execution time comparing to the naive GPU method.