As part of our collaboration with the NCSA Alliance on the development of commodity SuperClusters, we are heavily involved in benchamarking activity based on several major scientific applications. We plan to evaluate the performance of the configuration of machines by running these benchmarks and analyzing their performance on the machine configuration. Of particular interest is the CACTUS Computational Toolkit for Relativistic Astrophysics applications. These jobs can easily use 80+ Gbytes of memory and generate a terabyte of data. Although the awarded machine will not immediately support such large applications (not enough memory or disk capacity), we would initially develop techniques to obtain realtime snapshots that could be used to computationally steer the calculation as well as visualize the data. We would seek out opportunities such as an NSF CISE ACR award on Visualization And Advanced Data Management to enhance the memory and data storage to enable us to develop parallel input/output capabilities to handle such large amounts of data. The background for such an effort is that the Cactus toolkit has recently been ported to the RoadRunner Supercluster and has shown excellent performance compared to more costly SMP computing machines. Currently, Cactus runs on a variety of platforms including IBM/SP, SGI, T3E, NT, and Linux clusters. The planned machine allows us the opportunity to investigate which architectures are best suited for computation, visualization, and data movement and management on this large scientific application.