Broadcom Revolutionizes AI Networking Landscape with High-Performance 400G RoCE/RDMA Ethernet NICs

Cutting-Edge PCIe Technology Empowers Seamless Integration of Open Standards and Fabric Interconnects for AI Clusters

PALO ALTO, Calif., May 20, 2024 (GLOBE NEWSWIRE) -- Broadcom Inc. (NASDAQ: AVGO) today announced its latest portfolio of highly-scalable, high-performing, low-power 400G PCIe Gen 5.0 Ethernet adapters to revolutionize the data center ecosystem. These latest products offer an enhanced, open, standards-based Ethernet NIC and switching solution to resolve connectivity bottlenecks as XPU bandwidth and cluster sizes grow rapidly in AI data centers.

“At Broadcom, we recognize the significance of fostering a power-efficient and highly connected data center for AI ecosystem,” said Jas Tremblay, vice president and general manager of the Data Center Solutions Group, Broadcom. “Broadcom is prioritizing open standards and fostering collaboration with industry leaders to deliver the most extensive selection of high-performance connectivity solutions for AI infrastructure. Our 400G PCIe Gen 5.0 Ethernet adapters yet again underscore our commitment to enable the network-centric AI infrastructure platform.”

As the industry’s first Ethernet adapter built with 5nm process technology, the new portfolio enables the most power and thermally efficient design in the market. Combined with the device’s ability to drive passive copper cables up to five meters or ultra-low power linear pluggable optics transceivers and Broadcom’s high-radix Ethernet switches, the adapter delivers higher rack density using mainstream air-cooling technology.

The Broadcom family of Ethernet adapters meets the growing demand for higher data transfer speeds required in future AI networks. Equipped with a third-generation RoCE pipeline, low-latency congestion control technology and innovative telemetry features, the portfolio is ideal for the high-bandwidth, high-stress network environment that characterizes AI infrastructure.

“As the industry races to deliver generative AI at scale, the immense volumes of data that must be processed to train LLMs require even larger server clusters. Scalable high bandwidth, low latency connectivity is critical for maximizing the performance of these AI clusters,” said Patrick Moorhead, CEO & chief analyst at Moor Insights and Strategy. “Ethernet presents a compelling case as the networking technology of choice for next generation AI workloads. The 400G NICs offered by Broadcom, built on its success in delivering Ethernet at scale, offers open connectivity at an attractive TCO for power-hungry AI applications.”