Ultra Ethernet Consortium Sets Sights on InfiniBand in Pursuit of High Performance Computing and AI

TL;DR:

  • Ultra Ethernet Consortium (UEC) launches as a new organization under the Linux Foundation.
  • UEC aims to enhance Ethernet for high-performance computing and AI, challenging InfiniBand.
  • Founding members include AMD, Arista, Broadcom, Cisco, Eviden, HPE, Intel, Meta, and Microsoft.
  • UEC plans to make improvements at various layers of the Ethernet stack.
  • UEC Transport, a new technology, aims to provide flexible and high-performance Ethernet transport.
  • Interoperability and compliance testing will be prioritized by the UEC.
  • The consortium leverages the Linux Foundation’s governance framework for effective collaboration.

Main AI News:

In a bid to propel the pervasive networking standard for the artificial intelligence (AI) era, the Ultra Ethernet Consortium (UEC) has officially launched today under the auspices of the Linux Foundation and its Joint Development Foundation initiative. The UEC’s primary objective is to surpass the existing capabilities of Ethernet, such as remote direct memory access (RDMA) and RDMA over Converged Ethernet (RoCE), by providing a distributed, high-performance, and lossless transport layer specifically optimized for AI and high-performance computing. This ambitious initiative directly challenges the rival transport protocol InfiniBand.

Leading technology companies, including AMD, Arista, Broadcom, Cisco, Eviden, HPE, Intel, Meta, and Microsoft, have joined forces as founding members of the UEC. Their collective expertise and commitment form a solid foundation for tackling the networking challenges brought about by AI and high-performance computing.

J Metz, the chair of the steering committee for the Ultra Ethernet Consortium and a technical director at AMD, elaborated on the motivation behind this groundbreaking endeavor. Metz stated, “We wanted to create an approach to solving some of the problems that high-performance computing and AI have been bringing into the networking realm.”

The UEC’s comprehensive plan encompasses multiple layers of the Ethernet stack, from the physical layer to the software layer. While acknowledging the significant role of the IEEE in Ethernet standards, Metz emphasized that the UEC’s focus extends beyond the physical transport layer traditionally addressed by the IEEE. The consortium aims to scrutinize all aspects essential for enhancing Ethernet’s performance and collaborate with relevant standards bodies and technical groups to implement these improvements effectively.

One of the UEC’s initial undertakings is the development of UEC Transport, a revolutionary technology designed to enhance Ethernet transport beyond the capabilities of RDMA. While RDMA will remain supported, UEC Transport seeks to offer a more flexible transport solution that does not necessitate a lossless network. This innovation will enable advanced features like multi-pathing and out-of-order packet delivery, which are crucial for AI workloads involving multiple parties.

UEC Transport is a new approach to the transportation layer that introduces semantic adjustments, congestion notification protocols, and security enhancements,” explained Metz.

To facilitate smooth progress and ensure interoperability, the UEC has been incorporated under the Linux Foundation’s Joint Development Foundation, leveraging an existing legal and governance framework. Uri Elzur, chair of the technical advisory committee for the UEC, emphasized the consortium’s commitment to openness and interoperability. Elzur stated, “To meet our goals, there are certain aspects that must change while preserving the strengths of the Ethernet environment. We will build a stack on top of that foundation.”

Conclusion:

The establishment of the Ultra Ethernet Consortium (UEC) marks a significant development in the market for high-performance computing and AI networking. By bringing together industry-leading companies, the UEC demonstrates a collective commitment to advancing Ethernet beyond its existing capabilities. This concerted effort to improve Ethernet’s performance, interoperability, and flexibility will have profound implications for the market, potentially challenging established protocols like InfiniBand. With the UEC’s focus on multi-layer enhancements and the development of UEC Transport, the consortium is poised to reshape the landscape of networking standards, catering specifically to the demanding requirements of high-performance computing and AI applications.

Source