Ethernet for Scalable AI Cluster Deployments

Presented byBroadcom Inc.

Ethernet has become the leading fabric of choice for AI cluster deployments at major hyper scale customers. AI models need a distributed computing approach to scale up and scale out. Ethernet has the ability to provide performance at high scale, low job completion time with load balancing and efficient congestion management, with lower power and cost.

In this presentation we will discuss the advances in Ethernet based fabric for AI workloads that has enabled key customers to build very large networks. We will also discuss how to improve network utilization, improve job completion time and lower the total cost of ownership, all while ensuring a standards based ecosystem.