Enabling, Enhancing, and Extending Petascale Computing for Science and Engineering

目前的奖项授予德克萨斯大学奥斯汀分校，以部署和支持Stampede，这是一个HPC Linux集群，初始峰值性能为10petaflop。该系统将拥有6,400台Dell Stallion服务器，这些服务器通过FDR Infiniband连接。每台服务器都将拥有基于英特尔®S的双处理器，即将推出的沙桥架构具有32 GB的内存。该系统还将包括基于英特尔®多集成核心(英特尔®MIC)架构的即将推出的英特尔即将推出的“Knight Corner”协处理器的预发行版本：利用x86指令集的高度并行的协处理器。Stapede还将提供128个用于远程可视化的下一代NVIDIA GPU、16个具有1TB共享内存的节点用于大型数据分析，以及一个具有14 PB存储空间的高性能文件系统用于数据密集型计算。所有组件都将与FDR InfiniBand网络集成，以实现极高的可扩展性。总而言之，Stampede的峰值性能将超过10千万亿次浮点运算和超过250TB的内存。基于英特尔®MIC架构的第二代协处理器将在上市时添加，从而将Stampede的峰值性能提高到至少15千万亿次。这项国家资源将于2013年初通过国家科学基金会网络基础设施提供，以支持科学和工程方面的基础研究，并将运作和支持四年。该项目将推进包括英特尔®MIC架构性能优化在内的千万亿级计算方法，并将开发数据密集型计算方面的新专业知识。该奖项将使计算和数据驱动的科学和工程项目中的1000多个项目能够促进各自领域的知识。

The current award is to The University of Texas at Austin to deploy and support Stampede, a HPC Linux cluster with an initial peak performance of 10 petaflops. The system will have 6,400 Dell Stallion servers, with the servers connected by FDR Infiniband. Each server will have dual processors based on Intel?s forthcoming Sandy Bridge architecture with 32 GB of memory. The system will also include a pre-release version of Intel's forthcoming "Knights Corner" co-processors based on the Intel® Many Integrated Core ( Intel® MIC) Architecture: highly parallel co-processors that utilize the x86 instruction set. Stampede will also offer 128 next-generation NVIDIA GPUs for remote visualization, 16 nodes with 1TB of shared memory for large data analysis, and a high-performance file system with 14 petabytes of storage for data-intensive computing. All components will be integrated with a FDR InfiniBand network for extreme scalability. Altogether, Stampede will have a peak performance of over 10 petaflops and over 250 terabytes of memory. Second generation co-processors based on the Intel® MIC Architecture will be added when they become available, increasing Stampede's peak performance to at least 15 petaflops. This national resource will be available in early 2013 through the NSF Cyberinfrastructure to enable basic research in science and engineering, and will be operated and supported for four years. The project will advance methods for petascale computing including Intel® MIC Architecture performance optimization, and will develop new expertise in data-intensive computing. The award will enable 1000+ projects in computational and data-driven science and engineering projects to advance knowledge in their fields.