Parallel, Modified Nodal Integral Method and Innovative Application Accelerators (such as FPGAs, GPUs) for Turbulence Modeling using LES and DNS
Award year: 2008-2009
Accurate simulation of next of generation of nuclear power plants, just like many other engineering systems that involve turbulent flows, will rely on accurate simulation of turbulence. While computationally efficient modeling of turbulence remains a challenge, two approaches "large eddy simulation (LES) and direct numerical simulation (DNS)" have been shown to capture the essence of turbulence well. However, both of these approaches are computationally very expensive due to grid size limitations associated with the conventional numerical discretization approaches used. There are two remedies to lessen the computational burden, or more accurately, to improve the usefulness of LES and DNS: 1) Development of improved discretization methods to allow larger grid size; and 2) Exploit the latest paradigms in computational science (like parallel computation, grid computation, innovative application accelerators such as FPGAs, GPUs etc). A two pronged attack at the turbulence problem using these two approaches will allow simulation of problems that are otherwise impossible (within reasonable computational times) to simulate. Parallelization of computational fluid dynamics (CFD) codes has now reached production stage. However, newer developments associated with innovative application accelerators such as FPGAs, GPUs and the Cell Broadband Engine processor have not yet been adequately exploited and their full potential is yet to be explored. For example, some algorithms implemented on FPGAs in lab tests have been shown to be 10x to 1,000x faster than the traditional parallel algorithms. It is for this reason that many high-performance computing applications are turning to innovative application accelerators. When application accelerators are combined with advanced numerical discretization schemes, the resulting improvements in CFD stand to push the frontier of turbulence simulation to the next level.