Accelerating Deep Learning Recommendation Models using Disaggregated Memories

Departments: Electrical & Computer Engineering
Time: Monday, December 2, 2024 - 3:00pm - 4:00pm
Type: Seminar Series
Presenter: Vijay Narayanan, The Pennsylvania State University
Room/Office: Dunham Lab room 514 or Zoom
Location:
Dunham Laboratory, Room 514
10 Hillhouse Avenue
New Haven, CT 06511
United States
See map: Google Maps

Monday, December 2 at 3:00pm ET

DL514 or Zoom (https://yale.zoom.us/j/98967808687)

Hosted by: Priya Panda

Abstract: 

Deep Learning Recommendation model (DLRM) workloads form a critical share of modern-day data centers. As the size and complexity of the models increases, the memory system becomes a key bottleneck. The emergence of CXL-based memories provides an opportunity to enhance both memory capacity and bandwidth in addressing this bottleneck. However, the increased latency associated with these CXL memories pose a challenge. In this talk, I will show case a coordinated combination of hardware and software innovations to support DLRM workloads.

Biography:

Vijay Narayanan is an Evan Pugh University Professor and Robert A Noll Chair of Computer Science and Engineering at The Pennsylvania State University. His research interests are in Computer Architecture, Embedded Systems and Design using emerging device and packaging technologies. He has received the 2021 IEEE Computer Society Edward J. McCuskey Award, 2021 IEEE Computer Society TCVLSI Distinguished Research Award, 2022 ACM SIGDA Distinguished Service Award and the 2020 Northeastern Association of Graduate Schools Geoffrey Marshall Mentoring Award. He is a Fellow of the National Academy of Inventors, IEEE, ACM and AAAS.