Assessment of a heterogeneous computing CFD code in wind farm simulations

Abstract

The use of heterogeneous architectures, based on CPU-GPU processors, has led to a significant increase in the performance of parallel computing applications. In recent years, this approach has been implemented in various computational fluid dynamics (CFD) codes to take advantage of the compute capability of the GPU graphics cards. The objective of this work is to assess the performance of a general purpose CFD open-source code in wind energy applications, running in a heterogeneous architecture. To this aim, a numerical wind turbine model was migrated from a CPU-based Fortran program to the CFD code. Several timing tests were performed on a local computing station, while running simulations of well-documented wind tunnel experiments. The results obtained show a significant reduction in computational time and resource required, indicating a great potential of the GPU-accelerated CFD code to be used in large wind farms simulations or in real-time applications.