⚡ 本页包含 AI 生成的分析内容，仅供参考
本文提出了Spyre，一个针对企业推理工作负载优化的可扩展AI加速器，通过专用架构设计和高效数据流管理，解决了大规模部署中的性能与能效瓶颈。
Michael Guillorn1, Sandra Woodward3, JohnDavid Lancaster1, Josh Hursey4, Kyu-hyoun Kim1, Alberto Mannari5, Amrit Nagarajan1, Ananda Samajdar1, Bahman Hekmatshoartabari1, Bob Galbraith3, Ching Zhou1, Dave Satterfield1, Greg Still6, Gustavo Tellez1, Indira Nair1, Indu Masilamanai2, Jinwook Jung1, Kavi Randhawa2, Marcel Schaal1, Martin Lutz1, Paul Crumley1, Philip Jacob1, Prasanth Chatarasi1, Radhika Jain1, Saekyu Lee1, Sanchari Sen1, Sarada Krithivasan1, Scot Rider1, Shubham Jain1, Siyu Koswatta1, Thomas Roewer1, Tom Gooding3, Victor Ferrari7, Vidhi Zalani1, Zhibin Ren1, Kevin Reick2, Lisa Maurice8, Chris Gonzalez9, Chris Catalino2, Ryan Nett2, Pong-Fei Lu1, Rob Senger1, Leland Chang1 IBM Research, Yorktown Heights, NY, 2IBM Infrastructure, Austin, TX, 3IBM Infrastructure, Rochester, MN, 4IBM Research, Rochester, MN, 5IBM Research, Zurich, Switzerland, IBM Infrastructure, Research Triangle Park, NC, 7IBM Research, Sao Paulo, Brazil, 8IBM Research, Austin, TX, 9IBM Infrastructure, Yorktow
Matthew Cohen1, Monodeep Kar1, Swagath Venkataramani1, Viji Srinivasan1, Brian Veraa2, Matthew Ziegler1, Nianzheng Cao1, Ashish Ranjan1, Joel Silberman1,