Bill Gartner, SVP & GM of Optical Systems and Optical Business Unit at Cisco Systems, breaks down the complex architecture powering modern AI infrastructure. In this technical deep dive, Gartner explains how AI data centers require dramatically different network designs compared to traditional applications, with capacity demands reaching 500 times that of conventional wide area networks.
– The three distinct network domains that comprise AI infrastructure: scale-up, scale-out, and scale-across
– Why GPU connectivity within racks currently relies on copper interfaces and when optical solutions may take over
– How capacity requirements differ across each network layer, from intra-rack to inter-data center connections
– The role of coherent pluggable optics in enabling long-distance, high-capacity data center interconnection
– Cisco’s approach to addressing AI networking challenges, including the P200 silicon and deep-buffered routing technology
– The evolution path for optical technologies, from 400G to 1.6T and beyond
– Emerging questions around co-packaged optics (CPO) and near-packaged optics (NPO) for future scale-up networks
Gartner provides specific insights into Cisco’s deployment milestones and discusses the industry’s ongoing work to standardize next-generation optical interfaces for AI workloads.:
0:00:00 – AI Infrastructure Architecture: Three Core Network Elements
0:00:36 – Scale-Up Networks: GPU Connectivity Within Racks
0:00:57 – Scale-Out Networks: Expanding Beyond Single Rack Capacity
0:01:30 – Scale-Across Networks: Inter-Data Center Connectivity
0:02:48 – Cisco’s P200 Silicon and Deep-Buffered Routing Solutions
0:03:32 – Coherent Pluggable Optics: Deployment and Evolution
0:04:06 – Future of Scale-Up: Transitioning from Copper to Optical Solutions
Have a tech update that you want to brief us on? Contact [email protected]!
https://ngi.fyi/tech-cisco-billg-2
