Paper Title: Architecting Storage for AI/ML

Paper Abstract: High-performance deep-learning workloads have become common in nearly every data center. These GPU-based workflows place new demands, and new challenges, on the supporting storage infrastructure. This session delves into I/O characteristics common across the most prevalent AI/DL frameworks and the resulting impact on underlying storage technologies. Flash memory, storage interconnect, and system-level architecture are all discussed, as we describe what to think about as you architect storage solutions for AI/DL.

Paper Author: Steve McDowell, Sr Analyst, Moor Insights and Strategy

Author Bio: