Scaling AI Infrastructure: Optimizing Performance with High-Density 1.6 TbE Switching – Celestica Triple-T
As AI clusters scale toward hundreds of thousands of GPUs, the network fabric dictates the overall efficiency of the compute investment. Transitioning to 1.6TbE port switches enables architects to build a high-radix, 102.4Tbps foundation that flattens network hierarchy and eliminates additional hops.
This session demonstrates how 1.6TbE infrastructure optimizes AI workload performance at scale. We will evaluate the impact of 224G SerDes signaling and examine how support for Linear-drive Pluggable Optics (LPO) reduces per-link latency and cuts power consumption by up to 50%—a critical metric for sustaining the massive data bursts of large-model training.
Join the session to learn how a 102.4Tbps platform provides the flexibility required for modern AI data centers.
Arun Solleti is a seasoned technical product marketing professional with 20+ years of experience driving network infrastructure, cloud and enterprise solutions. Currently Director of Technical Product Marketing at Celestica, he previously drove innovation at Electronic Arts (Gen AI/cloud services), AWS (Amazon ECR), Extreme Networks (Campus Switching) and Cisco Systems. Arun has experience in products and solutions across enterprise markets such as Healthcare, Retail, Higher education etc. He holds an MBA from UNC Kenan-Flagler and MS in Computer Science from University of South Florida.
