AS COMPUTER ARCHITECTURE BECOMES MORE PARALLEL, NUMERICAL SIMULATIONS MUST FOLLOW SUIT AND EXPLOIT PARALLEL…

Public award description

AS COMPUTER ARCHITECTURE BECOMES MORE PARALLEL, NUMERICAL SIMULATIONS MUST FOLLOW SUIT AND EXPLOIT PARALLEL ALGORITHMS EFFECTIVELY. AN EXASCALE-CAPABLE COMPUTATIONAL FLUID DYNAMIC METHOD USING LATTICE BOLTZMANN METHOD (LBM) HAS BEEN PROPOSED TO SIMULATE COMPRESSIBLE FLOW. THE METHOD IS EMBARRASSINGLY PARALLEL WHICH ALLOWS THE METHOD TO FULLY UTILIZE PARALLEL ARCHITECTURES SUCH AS GRAPHICS PROCESSING UNITS (GPUS). THE METHOD IMPROVES UPON PREVIOUS METHODS AND ALLOWS FOR VARIABLE FLUID PROPERTIES INCLUDING SPECIFIC HEAT RATIO AND PRANDTL NUMBER. IN ADDITION, THE METHOD INCREASES THE TRADITIONAL MACH NUMBER LIMIT OF LBM FROM 0.2 TO 3.0 ALLOWING FOR THE METHOD TO SIMULATE TRANSONIC AND SUPERSONIC PHENOMENA. THE PHASE I PROJECT WILL INVESTIGATE THE ACCURACY AND SPEED OF THE METHOD WITH RESPECT TO EXISTING NASA SOLVERS INCLUDING NASA OVERFLOW AND NASA FUN3D. THE LBM SOLVER WILL BE WRITTEN IN SERIAL AND IN PARALLEL USING NVIDIA S CUDA TO ALLOW FOR GPU USE. FUTURE WORK IS DISCUSSED TO IMPROVE UPON THE METHOD AND TO INCORPORATE THE METHOD INTO NASA SOLVERS SUCH AS NASA LAVA AND NASA CART3D.