Summary:
- Meta is involved in the ESUN initiative, a collaboration with several tech giants to enhance networking technology for AI systems.
- ESUN focuses on open, standards-based Ethernet switching for scale-up networking, excluding proprietary technologies.
- Meta also introduced new data center networking innovations to improve flexibility, scalability, and efficiency in handling AI workloads.
Article:
The ESUN initiative, spearheaded by Meta, aims to revolutionize networking technology for AI systems through collaboration with industry leaders like AMD, Arista, Cisco, and more. This initiative focuses on advancing open, standards-based Ethernet switching to cater to the growing demands of scale-up networking for AI applications. By excluding proprietary technologies and application-layer solutions, ESUN aims to drive innovation in XPU network interfaces and Ethernet switch ASICs for scale-up networks.
In addition to ESUN, Meta unveiled three groundbreaking data center networking innovations during a recent event. These innovations include the evolution of Meta’s Disaggregated Scheduled Fabric (DSF) to support large AI clusters spanning entire data center buildings, the introduction of a new Non-Scheduled Fabric (NSF) architecture based on disaggregated Ethernet switches, and the incorporation of Minipack3N into Meta’s portfolio of OCP switches. These advancements are designed to enhance flexibility, scalability, and efficiency in managing AI workloads within Meta’s infrastructure.
Meta’s DSF, an open networking fabric, separates switch hardware, NICs, and other components to achieve a non-blocking fabric that interconnects thousands of XPUs. By utilizing scheduled fabric techniques like Virtual Output Queuing, Meta proactively avoids congestion and enhances performance for AI workloads. The evolution of DSF to a 2-stage architecture has enabled Meta to support AI clusters that span regions, meeting the increasing capacity and performance demands of Meta’s AI initiatives.
Overall, Meta’s involvement in the ESUN initiative and the introduction of new data center networking milestones highlight the company’s commitment to driving innovation in networking technology for AI systems. These developments not only enhance Meta’s infrastructure but also set new standards for scalability and efficiency in handling AI workloads.
