Description: Findings from our Distributed Deep Learning (DDL) workgroup on evaluating parallel deep learning frameworks using both "Onyx" Cray XC50 and "Hokulea" IBM Power 8 systems with Nvidia P100 GPUs are presented. Three areas are introduced within deep learning that necessitate distributed frameworks: (1) big data and the memory limitation of GPUs, (2) speeding up the training of deep neural networks via parallelization, and (3) hyperparameter optimization that requires running hundreds or thousands of alternative parameters.
Presenter(s): Dr. Wes Brewer - PET GDIT
Location: Webcast
Date & Time: February 4, 2020, 3:00p - 4:30p ET
Additional Notes: Part of the HPDA Seminar Series (2020).
Distribution Statement C. Distribution a authorized to U.S. Government Agencies and their contractors, Administrative or Operational Use, 04 February 2020. Other requests for this document shall be referred to the High Performance Computing Modernization Office, 3909 Halls Ferry Road, Vicksburg, MS 39180.