Pushing Infrastructure Boundaries: Memory Requirements in AI Data Centers

Explore a 20-minute keynote presentation from the SNIA Compute, Memory, and Storage Summit where Meta's Hardware Systems Technologist Manoj Wadekar examines the evolving landscape of hyperscale data centers. Discover how the shift from traditional CPU-centric platforms optimized for scale-out stateless applications to GPU-accelerated infrastructures supporting large-scale AI applications is challenging current system boundaries. Learn about the critical impact on accelerators, network architecture, memory systems, topologies, and rack power requirements as data centers adapt to support massive AI workloads. Gain insights into emerging memory use cases and understand the key areas of innovation needed to future-proof hyperscale infrastructure for next-generation AI applications.