The Difference between today’s Data Center networks and AI networks
Data Centers are designed for today’s workloads: typically Web, Database and Application with some specialized areas around Low-Latency and High-Performance Computing. AI is different and demands more bandwidth, power, and performance than what you provision today. You will learn about the differences in designing AI Networks compared to today’s networks: we will look at speeds, products, features and what choices customers have when looking to deploy AI models.
Product Update
Arista has a lot of new exciting products being released specifically for AI workloads. They include 800G support, visibility, telemetry, buffering, and feature sets designed specifically for the next generation of Data Center AI networks. We will discuss the next generation of AI-capable systems, including the 7060 series, 7800 series, and the new DES. These systems can be deployed standalone or in an AI leaf / spine cluster, with an additional deployment option of a single-hop scale-out design based on the 800G-optimized Arista 7700R series distributed Ethernet switch (DES).
AI Networking
We will discuss the ways in which AI traffic patterns are different from today’s traffic patterns: what do AI networks look like? What kind of bandwidth, performance, and features are required and which products are best suited to meet these needs. RoCEv2 - Congestion Control and Load Balancing RoCEv2 can protect applications from slowing down due to connectivity, congestion, and buffering issues. In this session we will provide an overview of PFC, ECN and DLB technologies and how they work together to ensure that Job Completion Time is as fast as possible
/Images%20(Marketing%20Only)/Icons/Events-time-date.png?width=85&height=85&name=Events-time-date.png){kind=icon}
/Images%20(Marketing%20Only)/Icons/Events-location-map.png?width=85&height=85&name=Events-location-map.png){kind=icon}