Scale-out Networking for AI and HPC Data Centers: Introducing Ultra Ethernet
Bldg. 320-109
About the talk: AI and HPC workloads present new challenges to the interconnection network for distributed computing systems, such as the need for higher scale, higher bandwidth density, multi-pathing, fast reaction to congestion, and inter-dependencies between the progress of individual flows. The Ultra Ethernet Consortium (UEC) was publicly launched in July 2023 (https://ultraethernet.org/). AMD was a founding member of the UEC alongside eight other major industry players: Arista, Broadcom, Cisco, Eviden (Atos), HPE, Intel, Meta, and Microsoft. There are now around 100 corporate members of the UEC. The mission of the UEC is to deliver the specification of an open complete architecture that optimizes Ethernet for high performance AI and HPC networking, exceeding the performance of today’s specialized technologies. The work focuses on functionality, performance, TCO, and developer and end-user friendliness, while minimizing changes to only those required and maintaining Ethernet interoperability. This talk will overview the background to the consortium and discuss its ongoing technical activities, building up to the first Ultra Ethernet specification that is expected towards the end of 2024.
About the speaker: Gordon Brebner contributes to AMD corporate vision and strategy in the general area of networking and distributed systems. As a research leader in AMD Research & Advanced Development, he has been investigating novel forms of programmability in high-performance, low-latency, and trustworthy communication networks, and translating results into successful products with an emphasis on ease of use through a particular focus on domain-specific languages and architectures. His current active focus is on the Ultra Ethernet Consortium (UEC), including being AMD’s voting representative on the Technical Advisory Committee. Gordon worked at the University of Edinburgh until moving to Silicon Valley in 2002, including being Head of the Department of Computer Science, and has a continuing connection there as an Honorary Professor in the School of Informatics.