BRIGE: Power-efficient easy-to-program 1000-core desktop supercomputer

BRIGE：高能效、易于编程的 1000 核台式超级计算机

• 批准号: 0926237
• 负责人: Tali Moreshet
• 金额: $ 8.37万
• 依托单位: Swarthmore College
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2009
• 资助国家: 美国
• 起止时间: 2009-08-15 至 2012-07-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=0926237&HistoricalAwards=false
• 关键词: BRIGE; Power; efficient; easy; program

IntroductionParallel computing in the form of chip multiprocessors has become an industry standard. The number of cores on a chip is expected to continue to increase in an effort to achieve continued performance growth while meeting thermal and power limits. However, the increasing power-performance scalability of multi-cores raises two main concerns. First, it is unclear how software will take advantage of the increasing hardware parallelism, and second, power consumption and thermal constraints are still a major design limitation.Intellectual MeritThis proposal focuses on a particular existing easy-to-program and easy-to-teach multi-core architecture, which provides a different approach to parallel programming, thus targeting the first concern above. The work intends to demonstrate the programmability of this architecture by incorporating it into parallel programming education at the college level. To address the second concern, the work aims to study the power consumption of this architecture and introduce novel power and thermal enhancements. Today?s microprocessor industry focuses on a single model for general-purpose multiprocessor computers. This proposal addresses the energy-efficiency of a new approach to parallel computing, which targets single task completion time. Ensuring that the new architecture meets power and thermal constraints will enable it to serve as a viable alternative approach to parallel computing. Broader ImpactsThe broader impact of this work is its potential to expose students to multicore architectures and parallel computing. This includes providing undergraduate students with research experience in energy-aware computer architecture and parallel programming, as well as incorporating these topics into existing courses and introducing new courses. A main goal of the proposed activities is to attract young people to engineering, as well as broaden participation of students in varied scientific areas, including students from currently underrepresented groups.

介绍并行计算的形式芯片多处理器已成为一个行业标准。芯片上的内核数量预计将继续增加，以实现持续的性能增长，同时满足热量和功率限制。 然而，多核的功率-性能可扩展性的增加引起了两个主要问题。 首先，目前还不清楚软件将如何利用日益增加的硬件并行性，第二，功耗和散热限制仍然是一个主要的设计limitation.Intellectual MeritThis建议侧重于一个特定的现有的易于编程和易于教的多核架构，它提供了一个不同的方法来并行编程，从而针对上述第一个问题。 这项工作的目的是证明这种架构的可编程性，将其纳入并行编程教育在大学一级。 为了解决第二个问题，这项工作的目的是研究这种架构的功耗，并引入新的功率和热增强。 今天吗？的微处理器工业集中于通用多处理器计算机的单一模型。 该建议提出了一种新的并行计算方法，其目标是单个任务完成时间的能源效率。 确保新架构满足功率和热约束将使其能够作为并行计算的可行替代方法。 更广泛的影响这项工作的更广泛的影响是它有可能让学生接触到多核架构和并行计算。 这包括为本科生提供能源感知计算机体系结构和并行编程方面的研究经验，以及将这些主题纳入现有课程并引入新课程。 拟议活动的一个主要目标是吸引年轻人学习工程学，并扩大学生对不同科学领域的参与，包括目前代表性不足的群体的学生。