Deep Learning Recommendation Models (DLRM) are widespread, account for a considerable data center footprint, and grow by more than 1.5x per year. With model size soon to be in terabytes range, leveraging Storage ClassMemory (SCM) for inference enables lower power consumption and cost. This paper evaluates the major challenges in extending the memory hierarchy to SCM for DLRM, and presents different techniques to improve performance through a Software Defined Memory. We show how underlying technologies such as Nand Flash and 3DXP differentiate, and relate to real world scenarios, enabling from 5% to 29% power savings.
@article{arxiv.2110.11489,
title = {Supporting Massive DLRM Inference Through Software Defined Memory},
author = {Ehsan K. Ardestani and Changkyu Kim and Seung Jae Lee and Luoshang Pan and Valmiki Rampersad and Jens Axboe and Banit Agrawal and Fuxun Yu and Ansha Yu and Trung Le and Hector Yuen and Shishir Juluri and Akshat Nanda and Manoj Wodekar and Dheevatsa Mudigere and Krishnakumar Nair and Maxim Naumov and Chris Peterson and Mikhail Smelyanskiy and Vijay Rao},
journal= {arXiv preprint arXiv:2110.11489},
year = {2021}
}