Innovative CFD Algorithm, Libraries & Python Frameworks for Hybrid-GPU Computing Architectures

The need for faster highly resolved solutions coupled with the advent of General Purpose Graphics Processing Unit (GPGPU) architectures and the development of GPGPU algorithms at the University of California, Davis present an opportunity that JMSI Inc. proposes to leverage by developing algorithmic and software solutions for GPGPUs in “Innovative CFD Algorithms, Libraries & Python Frameworks for Hybrid CPU-GPU Compute Architectures”. In the Phase 1 effort proposed herein, JMSI Inc. proposes to develop prototype algorithms for the core CFD primitives and matrix solvers that can be used with any CFD code through either libraries or a Python Interface Framework. The prototypes are based upon GPGPU research produced by the University COLLABORATORs, Profs. John Owens and Roger Davis, from the University of California, Davis. The MBFLO code by Prof. Davis will be used to demonstrate the core CFD primitives while a development implicit code will be used to demonstrate the use of implicit time stepping algorithms by Owens that are amenable to GPGPU architectures. BENEFIT: Accelerated CFD solutions at lower cost. Open source GPGPU libraries and Python Components that can be used with any CFD solver.