《2016 Rigel A 1024-Core Single-Chip Accelerator Architecture》.pdf

《2016 Rigel A 1024-Core Single-Chip Accelerator Architecture》.pdf

  1. 1、本文档共12页,可阅读全部内容。
  2. 2、原创力文档(book118)网站文档一经付费(服务费),不意味着购买了该文档的版权,仅供个人/单位学习、研究之用,不得用于商业用途,未经授权,严禁复制、发行、汇编、翻译或者网络传播等,侵权必究。
  3. 3、本站所有内容均由合作方或网友上传,本站不对文档的完整性、权威性及其观点立场正确性做任何保证或承诺!文档内容仅供研究参考,付费前请自行鉴别。如您付费,意味着您自己接受本站规则且自行承担风险,本站不退款、不进行额外附加服务;查看《如何避免下载的几个坑》。如果您已付费下载过本站文档,您可以点击 这里二次下载
  4. 4、如文档侵犯商业秘密、侵犯著作权、侵犯人身权等,请点击“版权申诉”(推荐),也可以打举报电话:400-050-0827(电话支持时间:9:00-18:30)。
查看更多
《2016 Rigel A 1024-Core Single-Chip Accelerator Architecture》.pdf

RIGEL: A 1,024-CORE SINGLE-CHIP ACCELERATOR ARCHITECTURE RIGEL IS A SINGLE-CHIP ACCELERATOR ARCHITECTURE WITH 1,024 INDEPENDENT PROCESSING CORES TARGETED AT A BROAD CLASS OF DATA- AND TASK-PARALLEL COMPUTATION. THIS ARTICLE DISCUSSES RIGEL’S MOTIVATION, EVALUATES ITS PERFORMANCE SCALABILITY AS WELL AS POWER AND AREA REQUIREMENTS , AND EXPLORES MEMORY SYSTEMS IN THE CONTEXT OF 1,024-CORE SINGLE-CHIP ACCELERATORS. THE AUTHORS ALSO CONSIDER FUTURE OPPORTUNITIES AND CHALLENGES FOR LARGE-SCALE DESIGNS. Increasing demand for perfor- such as control and instruction fetch across mance on data-intensive parallel workloads many processing elements. However, when has driven the design of throughput-oriented applications don’t naturally map to the parallel compute accelerators. For this work, SIMD execution model, programmers must we consider programmable accelerators in adapt their algorithms or suffer reduced effi- Daniel R. Johnson contrast to fixed-function or hardwired ciency. SIMD then limits the scope of appli- application-specific accelerator units. Cur- cations that can achieve the hardware’s peak Matthew R. Johnson rent programma

文档评论(0)

wgvi + 关注
实名认证
内容提供者

该用户很懒,什么也没介绍

1亿VIP精品文档

相关文档