Future of AI systems with Networked Memory Tiering (Netpreme co-located event @ OSDI 2026)
About This Event
About Event HBM capacity is scaling slowly. However, model size keep growing, context windows keep expanding, agentic systems keep getting more advanced. When facing memory capacity shortage, offloading to CPU memory has its own bottleneck: GPU stalls, slow model/context/session swapping, and unpredictable tail latency, etc. The current memory hierarchy is facing challenge. There is a need for a new memory tier that holds the "warm" data for the emerging agentic workloads. So what comes next? Netpreme is hosting an interdisciplinary panel featuring renowned researchers, seasoned industry practitioners, and visionary investors on the forefront of networked memory tiering and Memory Processing Units (MPUs) for next-generation AI infrastructure. Panelist: Panelists to be revealed — stay tuned Agenda: 8:00PM — Welcome and Refreshments 8:10PM — Lightning Talks, Panel and Q&A 8:50PM — Mingling and Networking Who this is for Systems researchers, ML infrastructure engineers, and anyone thinking seriously about memory bottlenecks in AI — whether you're building training clusters,…
See the rest of the description and register on Luma.