Overview of GPU Suitability and Progress of CFD Applications, and an Introduction of OpenACC for Directives-Based GPU Acceleration

Speakers: Stan Posey and Jeff Larkin, NVIDIA Corporation
April 21, 2015

Presentation

Abstract

Part 1: Current trends in high performance computing include the use of graphics processing units (GPUs) as massively-parallel co-processors to CPUs in order to accelerate numerical operations common to computational fluid dynamics (CFD) solvers. This first part of the talk will examine GPU parallelization strategies for various CFD methods and will provide examples of current implementations for research CFD software including FUN3D from NASA. To increase adoption of GPUs for commercial CFD, NVIDIA developed a library of multi-level solvers that includes algebraic multigrid (AMG) and others common to linear solvers. This includes parallelization of all phases of AMG including hierarchy construction and ILU factorization and solve. Examples will be provided that demonstrate the use of the library, AmgX for commercially deployed CFD software ANSYS Fluent and OpenFOAM.

Part 2: The second part of the talk will focus on OpenACC, which is a collection of compiler directives to specify loops and regions of parallel code. OpenACC permits use of standard C, C++ and Fortran for these parallel regions that can be offloaded from a host CPU to an attached accelerator. OpenACC is designed for portability across operating systems, host CPUs, and a wide range of accelerators including GPUs. A brief introduction to OpenACC will be provided, along with the basics on utilization, and current features and future directions.

Biography

Stan Posey joined NVIDIA in 2009 with more than 20 years of experience in the HPC industry. Posey began his career as an engineering analyst in applied CFD at the U.S. DOE Oak Ridge National Laboratory, and mechanical CAE consultant CD-adapco. However, limitations posed by meshing software of those days along with excessive compute times, motivated a transition to the vendor community to aid in improvements of CFD solutions. Posey held technical roles in CFD applications engineering and industry development during 17 years at SGI, and contributed to advances in parallel I/O for HPC application software during 3 years at Panasas. At NVIDIA, Posey is an HPC Program Manager where he leads the company’s strategy on HPC solutions for the domains of CFD and Earth system modeling. In this role, Posey works with engineers and scientists around the world to implement GPU acceleration that will pave the way towards exascale computing. Posey graduated with M.Sc. and B.Sc. degrees in Mechanical Engineering from the University of Tennessee Knoxville, and is a member of the AIAA who serves on the AIAA Computer Systems Technical Committee.

Jeff Larkin joined NVIDIA in 2013 as an HPC application engineer in the Developer Technology Group. In this role, he investigates and implements tuning and optimizations that enable GPU acceleration of end-user application software. In addition, Larkin serves as NVIDIA representative to the company’s membership in the OpenACC standards organization, and the OpenMP Architecture Review Board (ARB). Prior to NVIDIA, Larkin was an applications engineer at Cray for 7 years in the Cray Supercomputing Center of Excellence, at the U.S. DOE Oak Ridge National Laboratory. Larkin received a M.Sc. in Computer Science from the University of Tennessee Knoxville, and B.Sc. in Computer Science from Furman University.