Nokia is supporting the Ultra Ethernet Consortium (UEC) in its mission to bring together companies to develop Ethernet specifications and software APIs that could empower AI and HPC environments.
Nokia will bring its technology leadership in building super scalable, low latency deterministic networks for data centres and data centre interconnect (DCI) to the definition of the Ultra Ethernet standard being developed by the UEC.
Ethernet and the rise of AI
This is against the backdrop of AI driving the need for expanded computing resources and driving changes to networks, especially those connecting computing clusters for HPC used in AI training and inference. The near-universal protocol for data networks, Ethernet is an ideal place to start for meeting the performance needs of AI. Nokia has pledged to support the UEC in developing new standards, best practices and architectures for specialised AI data centre networks.
Nokia will bring its expertise to bear on defining Ultra Ethernet, leveraging its understanding of web, enterprise and service provider networks by contributing to several UEC working groups to help ensure that Ultra Ethernet will meet the needs of all its customers.
J Metz, Chair of the Ultra Ethernet Consortium, says: “The UEC welcomes Nokia’s deep experience and expertise in building some of the world’s highest-performing networks. Together with a broad ecosystem of consortium partners, our goal is to provide an open, interoperable standard that allows networks to be upgraded easily and inexpensively to meet the needs of AI and HPC applications. With strong partners like Nokia, we know the UEC will drive the kind of standard to ensure broad AI adoption.”
Vach Kompella, Head of Nokia’s IP business, says: “Generative AI and ML are revolutionising every aspect of our customers’ businesses. Our networking solutions advance their ability to deploy AI quickly and pursue exciting new use cases and applications. With our extensive footprint in communications, enterprise and webscale networks, we are focused on making Ultra Ethernet a highly compatible, low cost and interoperable part of future AI and HPC application stacks.”
