In-Datacenter Performance Analysis of a Tensor Processing Unit

Autor: Alek Jaworski, Suresh Bhatia, Kieran Miller, Rahul Nagarajan, Amir Salek, Gordon MacKean, Jeffrey Dean, Dan Steinberg, Sarah Bates, Matt Ross, Rick Boyle, Walter Wang, Mark Omernick, Albert T. Borchers, Narayana Penukonda, Ray Ni, Bo Tian, Diemthu Le, David A. Patterson, Aaron Jaffey, Ben Gelb, Andy Swing, Khaitan Harshit, Andrew Everett Phelps, Christopher Aaron Clark, Robert Hundt, Gregory Michael Thorson, Gregory Sizikov, Zhuyuan Liu, Michael J. Daley, Kathy Nix, Andy Koch, Horia Toma, Alexander Kaplan, C. Richard Ho, Steve Lacy, Maire Mahony, Nan Boden, Chris Severn, Rajendra Gottipati, Emad Samadiani, Adriana Maggiore, Norman P. Jouppi, Richard Walter, Mercedes Tan, Doe Hyun Yoon, Vijay K. Vasudevan, Jonathan Ross, Erick Tuttle, Doug Hogberg, Raminder Bajwa, Jed Souter, James Law, Robert Hagmann, William John Gulland, Ravi Narayanaswami, Jeremy Coriell, Naveen Kumar, Chris Leary, Tara Vazir Ghaemmaghami, Pierre-luc Cantin, Matt Dau, D. Hurt, Matthew Snelham, Julian Ibarz, Daniel Killebrew, John Hu, James Laudon, Cliff Young, Thomas Norrie, Kyle Lucke, Gaurav Agrawal, Clifford Chao, Nishant Patil, Alan Lundin, Eric Wilcox

Publikováno v: ISCA

Many architects believe that major improvements in cost-energy-performance must now come from domain-specific hardware. This paper evaluates a custom ASIC---called a Tensor Processing Unit (TPU) --- deployed in datacenters since 2015 that accelerates

Externí odkaz: https://explore.openaire.eu/search/publication?articleId=doi_dedup___::31a1493cafb65bd40f89e933605805fd
https://doi.org/10.1145/3140659.3080246