Reconfigurable Computing for High Performance Computing Computational Science

Parallel computing systems with thousands of processors are now common and more affordable due to the focus on clustered commodity processors. However, both market and physical factors are converging that will limit the performance of these systems in the future. Hardware advances over the past several decades have been empirically observed with remarkable precision to obey Moore's law, predicting an increase in transistor density by about a factor of two every eighteen months. Maintaining these improvements has become problematic as power dissipation and other size-limiting factors become more pronounced at smaller feature size. Reconfigurable computing, or heterogeneous computing, is offering hope to the scientific computing community as a way to provide continued growth in computing capability. This paper discusses some of the hardware and software associated with this new technology. It also provides a discussion on the overall state of the technology for use by computational scientists. This is done by exploring a sample problem related to bit- and integer-based applications.