Algorithm Development for Reconfigurable Computing Architectures

The demands of high performance engineering calculations have pushed modern computing architectures based on scalar multiprocessors to their very limits. Larger systems, unstructured grids, multi-scale and multi-phase requirements have all worked to dramatically increase problem complexity. This has created an urgent need for faster, more capable processing techniques. Implementing software algorithms in hardware is one approach to a significant reduction in processing time, however, the specialized nature of custom solutions makes them expensive, inflexible and useful for only the most critical tasks. Field Programmable Gate Arrays (FPGAs) offer the speed of dedicated hardware with the programmable flexibility of software. FPGAs may be inserted into workstations and clusters to provide very effective acceleration. The state of the art suffers from development environments that are ill-suited to general purpose programming and the lack of low cost hardware that is designed to this task. In this project we propose to develop and demonstrate innovative algorithms and computationally effective solutions to problems of engineering physics for the solution of differential equations and the simulation of physics via stochastic techniques.