Missile Plume Simulation Improvements Using GPU Chemical Kinetics Coprocessors

High-fidelity missile plume flowfield simulations of MDA interest require use of detailed chemical kinetic mechanisms, which significantly improve IR/UV/RCS/visible signature prediction but entail long solution runtimes for completion. These long runtimes result from the required iterative solution of large systems of stiff, non-linear chemical source terms at each CFD mesh point; this curtails their routine use for practical systems-level studies. The standard approach for improving runtime performance, using distributed massively parallel clusters, has limited speed-up capability since it relies on very fine-grained domain decompositions, which increase inter-processor communication overhead and introduce numerical boundary stiffness issues. Alternatively, Graphical Processor Units (GPUs) have several hardware features conducive to scientific computing applications and are better suited to directly solving chemical kinetics problems using innovative data parallel algorithms. In this manner, GPUs can be effectively programmed as "chemical reaction co-processors." Time involved with GPU data transfer and computation can coincide with tasks remaining on the CPU to effectively minimize wall-clock time dedicated to chemical specie source term evaluation and hence remove a major plume simulation bottlenecking issue.