This award is funded under the American Recovery and Reinvestment Act of 2009 (Public Law 111-5).
Studying the way molecules interact in proteins and polymers, or how stars interact dynamically in very dense clusters, requires very powerful and very fast computers. The capabilities of certain kinds of computer processors, called Graphical Processing Units (GPUs) have been expanding rapidly due to a large commercial market. They are now faster, and cheaper, than conventional Central Processing Units (CPUs) at the heart of most computers. Dr. Stephen McMillan and his colleagues at Drexel University are proposing to use these fast, efficient, and affordable CPUs to speed up the very complicated and lengthy computations needed in to model the behavior of large collections of stars or molecules. NSF's Major Research Instrumentation program is funding Dr. McMillan's work through its Division of Astronomical Sciences.
The goal of this project was to build and maintain a high-performance computer cluster ("Draco") accelerated by Graphics Processing Units, configured to address specific performance goals mandated by the research programs of the PIs. The system was delivered in January 2010 and has run with minimal unanticipated downtime for the last 4 years. It has nominal peak performance substantially exceeding 200 Tflop/s and total memory of 2.3 Tbyte. The large increase in computational power provided by Draco has led to breakthrough applications in astrophysics, biophysics and biomedical imaging. Specific scientific applications include: Astrophysics of galaxies and star forming regions. The cluster has supported simulations in galaxy dynamics, star formation, and star-cluster evolution. This includes implementation of one of the fastest galaxy simulation codes currently in existence. Microscopic modeling of molecular dynamics, including all-atom molecular simulations of biological molecules. Of particular importance is the first implementation of a novel and efficient approach to studying protein dynamics. High-performance image registration applications for the medical community, including highly data-parallel deformable image registration algorithms suitable for use on many-core processors such as the GPU. In addition, the cluster has been used as a development and testing platform for new software packages. Free and open source software is playing an increasing important role throughout society, and provides a common economic good by reducing duplicated effort, and advances science by promoting the open exchange of ideas. Using the hardware obtained under this grant, the PIs have developed and disseminated open-source software packages applicable to the domains of comutational astrophysics and biomedical imaging.
