A multi-block viscous flow solver based on GPU parallel methodology

• A compressible CFD code with latest numerical methods and turbulent models is presented and further assessed.
• We show our successful design of a highly parallel computation system based on GPU cards.
• We create 2 streams in order to achieve memory copying concurrently with kernel executing on GPU.
• High order numerical methodologies for RANS are established and successfully applied to complex aircraft simulations.
• Scale-Adaptive Simulation model is constructed for both industrial and academic researches.

A multi-block viscous flow solver for steady and unsteady turbulent flows based on GPU parallel methodology under the finite volume frame is presented in this paper. Both numerical accuracy and computational efficiency are concerned. Numerical flux scheme for all speeds SLAU is adopted because of its wide adaptability and strong robustness; high order reconstruction schemes like MLP and WENO are chosen to evaluate the inviscid terms while a set of fully conservative 4th-order central differencing schemes are utilized to deal with the viscous terms. Second-order temporal accuracy is forfeited for unsteady simulations by coupling DP-LUR with dual time-stepping strategy. Furthermore, heterogeneous multiple CPU + GPU coprocessing system is well established with CUDA and MPI methodology. Design details about GPU implementation are analyzed and discussed. Impressive speedup factor is achieved on our GPU platform compared with CPU indicating the bright feature of these algorithms. Numerical results of several complex configurations have demonstrated the validity and reliability for aerospace engineering applications.