Paper Title: New and Faster Connectivity for Storage 's 400, 800 Gigabit and beyond

Paper Abstract: Accelerating computing and AI models are computed using massive amounts of data to refine the model accuracy. Many times, using large files containing video and photo in multiple complex data formats. Data sets have jumped from being measured in GigaBytes to now multiple TeraBytes or as we say at NVIDIA - "Terror-Bytes"! Unlike the past, AI compute problems now are much larger than a single server, now spanning multiple servers and computing for sometimes months at a time. This means moving TeraBytes of data in and out of modeling servers hundreds of times and across large configurations of multiple compute servers at long distances. This is requiring new networking technologies for storage. • This presentation discusses: o 400G, 800G, 1.6T and 3.2Tb transceivers in InfiniBand and Ethernet in development o OSFP, QSFP-DD, QSFP112 and SFP112 transceiver form-factors o Single mode, multimode optics, new optical connectors, and fibers o Passive and active copper cables o 100G-PAM4 signaling and the 200G-PAM4 outlook

Paper Author: Brad Smith, Director of Marketing, Interconnects, NVIDIA, Networking, LinkX Interconnects

Author Bio: Brad Smith is the Director of Marketing for NVIDIA’s Link cables and transceiver line of InfiniBand and Ethernet products. Brad has over 30-years’ experience in data center technologies and networking, holds 3 optical patents with backgrounds at Intel Silicon Photonics, Mellanox networking, and AMD CPUs. Brad Has BSEE from University of Massachusetts, and MBA University of Phoenix