Paper Title: Design the Right Storage Systems to Accelerate Deep Learning

Paper Abstract: This talk presents an analysis of multiple storage technologies and architectures that can be used for deep learning processing acceleration. DL is often related to computing and memory bandwidth bottlenecks, where GPUs can bring a solution. Smart solutions can be designed by re-thinking the overall system architecture and using current and innovative storage technologies: targeting a data movement reduction and leading to a computing efficiency increase. You will learn: -architecture: how to deal with NVMe, NVDIMM and NVMe-oF for computational storage -applications: benefits for deep learning (training and inference)

Paper Author: Jerome Gaysse, Senior Technology and Market Analyst, Silinnov Consulting

Author Bio: Jerome Gaysse is senior technology & market analyst at Silinnov Consulting, a consultancy based on emerging data storage and processing technologies. He manages business growth through innovation, including product strategy and long term technology roadmap. In addition to his creativity, visionary and entrepreneurship skills, he has strong technical expertise in IP, NVM, FPGAs, and ASICs for the data center market. He successfully worked with major corporate companies, startups, and research institutes in both Europe and the USA. He is a regular presenter at international conferences such as those sponsored by IEEE and SNIA, as well as past Flash Memory Summits. Jerome earned an MSEE from the National Institute of Applied Sciences (INSA) in Lyon (France) with a semiconductor specialty.