Welcome Remarks
summary
Announces a dedicated backend network (RDMA) built separate from front-end services, designed to support tens of thousands of GPUs, exabytes of storage and terabits per second of traffic; claims AI workloads scale 1000x every 2–3 years and require lossless, low-latency behavior. Describes deploying dynamic traffic engineering to place jobs and avoid congestion, host-level tuning of NIC/GPU/CPU bottlenecks, and a suite of monitoring and troubleshooting tools; two live Q&A sessions and recordings follow.
AI-written summary · about
related talks
06mOpening Remarks
13mLive Q&A Session #2 – Shashi Gandham
31mLive Q+A 1
10mLightning Talks Introduction - AI for NI
31mLive Panel: Scale-Out Networking for AI Supercomputers
12mPreview: Networking
18mArchitecting Multi-tenant Data-center Networks for GPU Customers
19mAI/ML Networking Challenges: The Fast and the Finicky!