Acceleration of Turbulent Flow Simulations with Intel Xeon Phi(TM) Manycore Processors

Enhancing the performance of turbulent flow simulations is important as the size of simulations grows with higher Reynolds number. We discuss the performance of our in-house turbulent flow simulation solver, named as DNS-TBL (Direct Numerical Simulation: Turbulent Boundary Layer), on the Intel Xeon Phi™ manycore processors. With bootable Knights Landing processors, the DNS-TBL solver shows excellent parallel scalability and, in particular, shows a 1.6 times better performance in solver time than the original CPU-based version. Current intermediate results serve as a practical case-study which directly shows how much turbulent flow simulations can be accelerated on manycore processors, providing a good reference for the parallelization and optimization schemes in the filed of computational fluid dynamics.