First-Generation Inference Accelerator Deployment at Facebook

Autor: Anderson, Michael, Chen, Benny, Chen, Stephen, Deng, Summer, Fix, Jordan, Gschwind, Michael, Kalaiah, Aravind, Kim, Changkyu, Lee, Jaewon, Liang, Jason, Liu, Haixin, Lu, Yinghai, Montgomery, Jack, Moorthy, Arun, Nadathur, Satish, Naghshineh, Sam, Nayak, Avinash, Park, Jongsoo, Petersen, Chris, Schatz, Martin, Sundaram, Narayanan, Tang, Bangsheng, Tang, Peter, Yang, Amy, Yu, Jiecao, Yuen, Hector, Zhang, Ying, Anbudurai, Aravind, Balan, Vandana, Bojja, Harsha, Boyd, Joe, Breitbach, Matthew, Caldato, Claudio, Calvo, Anna, Catron, Garret, Chandwani, Sneh, Christeas, Panos, Cottel, Brad, Coutinho, Brian, Dalli, Arun, Dhanotia, Abhishek, Duncan, Oniel, Dzhabarov, Roman, Elmir, Simon, Fu, Chunli, Fu, Wenyin, Fulthorp, Michael, Gangidi, Adi, Gibson, Nick, Gordon, Sean, Hernandez, Beatriz Padilla, Ho, Daniel, Huang, Yu-Cheng, Johansson, Olof, Juluri, Shishir, Kanaujia, Shobhit, Kesarkar, Manali, Killinger, Jonathan, Kim, Ben, Kulkarni, Rohan, Lele, Meghan, Li, Huayu, Li, Huamin, Li, Yueming, Liu, Cynthia, Liu, Jerry, Maher, Bert, Mallipedi, Chandra, Mangla, Seema, Matam, Kiran Kumar, Mehta, Jubin, Mehta, Shobhit, Mitchell, Christopher, Muthiah, Bharath, Nagarkatte, Nitin, Narasimha, Ashwin, Nguyen, Bernard, Ortiz, Thiara, Padmanabha, Soumya, Pan, Deng, Poojary, Ashwin, Ye, Qi, Raginel, Olivier, Rajagopal, Dwarak, Rice, Tristan, Ross, Craig, Rotem, Nadav, Russ, Scott, Shah, Kushal, Shan, Baohua, Shen, Hao, Shetty, Pavan, Skandakumaran, Krish, Srinivasan, Kutta, Sumbaly, Roshan, Tauberg, Michael, Tzur, Mor, Verma, Sidharth, Wang, Hao, Wang, Man, Wei, Ben, Xia, Alex, Xu, Chenyu, Yang, Martin, Zhang, Kai, Zhang, Ruoxi, Zhao, Ming, Zhao, Whitney, Zhu, Rui, Mathews, Ajit, Qiao, Lin, Smelyanskiy, Misha, Jia, Bill, Rao, Vijay

In this paper, we provide a deep dive into the deployment of inference accelerators at Facebook. Many of our ML workloads have unique characteristics, such as sparse memory accesses, large model sizes, as well as high compute, memory and network band

Externí odkaz: http://arxiv.org/abs/2107.04140

Deep Learning Inference in Facebook Data Centers: Characterization, Performance Optimizations and Hardware Implications

Autor: Park, Jongsoo, Naumov, Maxim, Basu, Protonu, Deng, Summer, Kalaiah, Aravind, Khudia, Daya, Law, James, Malani, Parth, Malevich, Andrey, Nadathur, Satish, Pino, Juan, Schatz, Martin, Sidorov, Alexander, Sivakumar, Viswanath, Tulloch, Andrew, Wang, Xiaodong, Wu, Yiming, Yuen, Hector, Diril, Utku, Dzhulgakov, Dmytro, Hazelwood, Kim, Jia, Bill, Jia, Yangqing, Qiao, Lin, Rao, Vijay, Rotem, Nadav, Yoo, Sungjoo, Smelyanskiy, Mikhail

The application of deep learning techniques resulted in remarkable improvement of machine learning models. In this paper provides detailed characterizations of deep learning models used in many Facebook social network services. We present computation

Externí odkaz: http://arxiv.org/abs/1811.09886

Tento výsledek nelze pro nepřihlášené uživatele zobrazit.
K zobrazení výsledku je třeba se přihlásit.

Efficient Approximation Algorithms for Weighted $b$-Matching

Autor: Fredrik Manne, Arif O. Khan, Mahantesh Halappanavar, Pradeep Dubey, Md. Mostofa Ali Patwary, Narayanan Sundaram, Nadathur Satish, Alex Pothen

Publikováno v: SIAM Journal on Scientific Computing. 38:S593-S619

We describe a half-approximation algorithm, $b$-Suitor, for computing a $b$-Matching of maximum weight in a graph with weights on the edges. $b$-Matching is a generalization of the well-known Matching problem in graphs, where the objective is to choo

Externí odkaz: https://explore.openaire.eu/search/publication?articleId=doi_________::81ffd668bad15a3b82836811b4153878
https://doi.org/10.1137/15m1026304

Banshee

Autor: Christopher J. Hughes, Xiangyao Yu, Srinivas Devadas, Nadathur Satish, Onur Mutlu

Publikováno v: MICRO
MIT web domain

Putting the DRAM on the same package with a processor enables several times higher memory bandwidth than conventional off-package DRAM. Yet, the latency of in-package DRAM is not appreciably lower than that of off-package DRAM. A promising use of in-

Externí odkaz: https://explore.openaire.eu/search/publication?articleId=doi_dedup___::57e65ebd5916727f236df6e6a627a04f
https://doi.org/10.1145/3123939.3124555