Accelerating Fast Fluid Dynamics Simulations with OpenACC
Understanding turbulent fluid flows lies at the heart of such versatile domains as climate, health, and energy. Modern methods used in flow simulations have yet to reach their efficiency limits in providing fast, high-resolution predictions. To answer outstanding questions about the nature of turbulence requires massive computing power and innovative methods designed with computational efficiency at their core.
CaNS, an open-source fast finite difference incompressible Navier-Stokes solver, is a modern Fortran code that focuses on high performance and hardware adaptivity. By porting the application to GPUs, CaNS has achieved remarkable performance and is now able to simulate wall turbulence at unprecedented Reynolds numbers (i.e., high flow speeds).
“Direct interaction with OpenACC’s domain-specific GPU-porting experts and community groups was key to achieving the high performance we sought,” shares Dr. Costa from TU Delft in his presentation at the Open Accelerated Computing Summit.
Watch Dr. Costa detail the guidelines and best practices of GPU porting the CaNS application.