Designed to support more than 1 million accelerator chips in AI data centers, the Tomahawk 6 Ethernet switch series is being called a ‘turning point in AI infrastructure design’ by Broadcom, which says deployments with more than 100,000 accelerator chips are being planned.
Broadcom is calling its newly launched 102.4-Tbps Tomahawk 6 Ethernet switch series a “breakthrough” and a “turning point in AI infrastructure design” that is “poised to make a rapid and dramatic impact on the deployment” of AI data centers.
The IT infrastructure giant announced on Tuesday that it has begun shipping the Tomahawk 6 switch series, representing what it called the “world’s first 102.4 Tbps of switching capacity in a single chip, double the bandwidth of any Ethernet switch” available now.
[Related: Intel Spin-Off: Our InfiniBand Alternative For AI Data Centers Has A ‘Devastatingly Good’ Edge]
“Tomahawk 6 is not just an upgrade—it’s a breakthrough,” said Ram Velaga, senior vice president and general manager of Broadcom’s Core Switching Group, in a statement. “It marks a turning point in AI infrastructure design, combining the highest bandwidth, power efficiency, and adaptive routing features for scale-up and scale-out networks into one platform.”
The Tomahawk 6 announcement was lauded by major IT channel vendors such as Juniper Networks, Arista Networks and AMD.
“We are excited to leverage its low-power, high- radix, 1.6-Tbps port speeds, and advanced packet processing capabilities for AI-aware routing, based on open Ethernet standards,” said Hardev Singh, general manager of cloud titans and AI at Arista, in a statement.
Velega called demand from customers and partners “unprecedented,” with Broadcom noting there are “multiple deployments” planned with more than 100,000 accelerator chips “using Tomahawk 6 for both the scale-out and scale-up interconnect.”
“Tomahawk 6 is poised to make a rapid and dramatic impact on the deployment of large AI clusters,” Velega said.
Broadcom said Tomahawk 6, which is compliant with specifications by the widely supported Ultra Ethernet Consortium, features the “industry’s most comprehensive set of AI routing features and interconnect options,” which will allow the switch series to scale AI data centers to nodes with more than 1 million accelerator chips.
The switch chip architecture also enables “unified networks for AI training and inference at unprecedented scale,” thanks to capabilities like advanced telemetry, dynamic congestion control and rapid failure detection as part of Broadcom’s Cognitive Routing 2.0 feature.
Thanks to its use of “industry-leading” 200-Gbps SerDes technology, Tomahawk 6 “provides the longest reach for passive copper interconnect, enabling high-efficiency, low-latency system design with the highest reliability and lowest total cost of ownership,” according to Broadcom.
Tomahawk 6 also supports 1,024 100-Gbps SerDes ports on a single chip, “enabling AI clusters with extended copper reach and efficient use of XPUs with native 100G interfaces,” the company said.
Broadcom said the new switch chip family will provide customers with an option for co-packaged optics to enable connectivity at the “lowest power and latency while reducing link flaps and improving long-term reliability,” which it called essential for hyperscalers.
