Scalable Aero-Load and Aero-Elasticity Solvers for Massively Parallel Heterogeneous Computing Architectures

ABSTRACT: We propose to apply modern massively multicore processors to a key problem area of interest to the Air Force: multiphysics computational fluid dynamics (CFD) and computational structural dynamics (CSD) solvers. The capabilities of the CPU to solve these problems have been increasing steadily, but the CPU is still a general-purpose device designed to run diverse applications such as word processors and internet browsers - it is not a high performance device for scientific computing! One of the emerging technologies in high-performance computing (HPC) are graphics processing units (GPUs); driven by market leader NVIDIA, the GPU has become a highly respected platform for computing. Among the codes under consideration for this project are: CREATE-Kestrel, NSU3D, AVUS/Cobalt, and USM3D. Such codes are widely used at both the Air Force and in the commercial and government spaces as well. In this project, we will apply our expertise in the GPU computing field to a key set of multiphysics solvers in the CFD/CSD space for aerodynamic and aero-elastic analysis. The results will be a manyfold improvement in speed, and a reduction in cost as less hardware will be required to solve any given problem. BENEFIT: At the end of Phase II, the solvers will be deployable at the Air Force immediately and at other centers that use the same code. Expected customers are the Navy, NASA, Boeing, Lockheed, and Raytheon. Customer (Non-AF) revenue will support ongoing enhancements and upgrades in the Phase III period. This is especially true in the case of codes from the CREATE (Kestrel, Helios, etc) program, which are owned by the Department of Defense and yet expected to be used widely.