EAGER: An Energy-Driven Point Design Study for Exascale Computing

EAGER：针对百亿亿次计算的能量驱动点设计研究

• 批准号: 0936700
• 负责人: Stephen Keckler
• 金额: $ 19.99万
• 依托单位: University of Texas at Austin
• 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: 2009-09-01 至 2012-08-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=0936700&HistoricalAwards=false
• 关键词: EAGER; Energy; Driven; Point; Design

The NSF and other Federal agencies will require further understanding of concepts, methods, and means for Exascale computing systems before formulation of future programs in enabling technologies, architectures, and methodologies. NSF supports a targeted two-year program of investigation, The Exascale Point-design Study, to provide sufficient depth and detail of understanding of the interrelationships among possible enabling technologies, implementation architectures and operating system software, and programming interfaces and methodologies.Working in consultation with LSU, USC, UIUC, UD, and Sandia National Laboratory (under DOE sponsorship), UT-Austin team focuses on energy efficiency as a key driver for Exascale systems. The PIs plan to investigate processor, memory system, and interconnect architectures that enable the exploitation of temporal and spatial locality, even when traditional memory hierarchies perform poorly. They plan to further investigate the use of hardware specialization as a means to accelerate critical components of applications in an energy-efficient fashion. UT-Austin plans to devise an Exascale architecture derived from prior research conducted under NSF, DOE, and DARPA sponsorship, reflecting some of the best-known principles and practices envisioned for hardware and software at the Exaflops scale.

在制定支持技术、架构和方法的未来计划之前，美国国家科学基金会和其他联邦机构将需要进一步了解百亿亿次计算系统的概念、方法和手段。 NSF 支持一项为期两年的有针对性的调查计划，即百兆级点设计研究，以提供对可能的支持技术、实施架构和操作系统软件以及编程接口和方法之间相互关系的足够深度和细节的理解。 UT-Austin 团队与 LSU、USC、UIUC、UD 和桑迪亚国家实验室（在 DOE 赞助下）协商，重点关注能源领域 效率是百亿亿次级系统的关键驱动因素。 PI 计划研究处理器、内存系统和互连架构，即使传统内存层次结构性能不佳，也能利用时间和空间局部性。他们计划进一步研究使用硬件专业化作为以节能方式加速应用程序关键组件的手段。德州大学奥斯汀分校计划设计一种基于先前在 NSF、DOE 和 DARPA 赞助下进行的研究的 Exascale 架构，反映了针对 Exaflops 规模的硬件和软件设想的一些最著名的原则和实践。