Broadcom announced it started shipping the Tomahawk 5 Ethernet switch/router chips in production volume.
“Taking Tomahawk 5 to production in seven months after the product was sampled is a testament to Broadcom’s execution quality,” said Ram Velaga, senior vice president and general manager, Core Switching Group, Broadcom.
Tomahawk 5 provides a performance boost for AI/ML infrastructure, delivering 51.2 terabits/second of switching capacity, twice that of any other switch on the market. Tomahawk 5 enables the fastest data transfer between AI/ML endpoints.
The high radix of Tomahawk 5 enables single-hop connectivity between 256 high-performance AI/ML accelerators, each having 200Gbps of network bandwidth. This results in the fastest completion time for AI training and inference jobs, including for today’s generative AI models.
Tomahawk 5’s Cognitive Routing ensures high network utilization. The Cognitive Routing suite includes innovative features such as Global Load Balancing which uses distributed inter-switch communication of congestion information to choose the best global path for packets through the network. Cognitive Routing also provides high network resiliency with extremely fast link-failover by detecting a failed link and re-directing traffic in less than 500ns.