High Performance System Acquisition: Enhancing the Petascale Computing Environment for Science and Engineering

The NSF's vision for Cyberinfrastructure in the 21st Century includes enabling sustained petascale computational and data-driven science and engineering through the deployment and support of a world-class High Performance Computing (HPC) environment. For the past decade the NSF has provided the open science and engineering community with state of the art HPC assets ranging from loosely coupled clusters, to large scale instruments with many thousands of computing cores communicating via fast interconnects. Previous solicitations, as exemplified by the multi-pronged Track Two acquisitions, have provided more than two petaflops (1015 floating point operations per second) of compute power on real applications, that consume large amounts of memory, and work with very large data sets. These resources have been made available through the TeraGrid, the world's largest, most powerful and comprehensive distributed cyberinfrastructure for open science. In addition to the Track Two acquisitions, the ongoing Track One program promises to deliver a petaflop of sustained power capable of tackling some of the most challenging scientific problems across multiple science and engineering domains. HPC Resource Providers - those organizations willing to acquire, deploy and operate HPC resources in service to the science and engineering research and education community - play a key role in the provision and support of a national Cyberinfrasructure. With this solicitation, the NSF requests proposals from organizations willing to serve as HPC Resource Providers within Extreme Digital (XD), the successor to TeraGrid, and who propose to acquire and deploy new, innovative petascale HPC systems and services.Competitive HPC systems will:Expand the range of data intensive computationally-challenging science and engineering applications that can be tackled with XD HPC services; Introduce a major new innovative capability component to science and engineering research communities:Provide an effective migration path to researchers scaling data and code beyond the campus level;Incorporate reliable, robust system software and services essential to optimal sustained performance; Efficiently provide a high degree of stability and usability by January, 2013; andComplement and leverage existing XD capabilities and services.Benchmarks will be a key factor in system selection. Two types of benchmarks are required: NSF provided benchmarks and proposer selected benchmarks. The NSF provided benchmarks, which are posted on NSF web-siteNSF 0605 are designed to capture the salient attributes of those science and engineering applications which will place the most stringent demands on the overall system to be provisioned. Proposer provided benchmarks should focus on the innovative aspect of the proposal .