Leadership-Class System Acquisition - Creating a Petascale Computing Environment for Science and Engineering

NSF’s goal for high performance computing (HPC) in the period 2006-2011 is to enable petascale science and engineering through the deployment and support of a world-class HPC environment comprising the most capable combination of HPC assets available to the academic community. The petascale HPC environment will enable investigations of computationally challenging problems that require computing systems capable of delivering sustained performance approaching 1015 floating point operations per second (petaflops) on real applications, that consume large amounts of memory, and/or that work with very large data sets. Among other things, researchers will be able to perform simulations that are intrinsically multi-scale or that involve the simultaneous interaction of multiple processes. HPC Resource Providers - those organizations willing to acquire, deploy and operate HPC systems in service to the broad science and engineering research and education community - play a key role in the provision and support of a national HPC environment. With this solicitation, NSF requests proposals from organizations, or groups of organizations, willing to serve as a petascale HPC Resource Provider, and who propose to acquire and deploy a new, state-of-the-art, petascale HPC system. A competitive, petascale HPC system will: *Enable researchers to work on a range of computationally-challenging science and engineering applications at the frontiers of research; *Incorporate reliable, robust system software essential to optimal sustained performance; *Provide a high degree of stability and usability; and, *Function as a community-driven resource that actively engages the research and education communities in petascale science and engineering. A robust and effective HPC acquisition process, driven by the requirements of the science and engineering research and education community, is one of the key elements of NSF’s HPC strategy. Accordingly, the desired capabilities of the system to be acquired are defined in terms of performance on model problems.