Broadcom has introduced the Tomahawk 6 chip, enabling Ethernet switches in data centers to achieve processing speeds of 102.4 terabits per second, crucial for large AI clusters. This chip addresses bandwidth bottlenecks in data-intensive workloads by utilizing Cognitive Routing 2.0 to manage traffic efficiently. Additionally, it offers co-packaged optics that reduce costs and power consumption by integrating transceivers directly into the switch. Capable of supporting extensive configurations, the Tomahawk 6 can manage clusters of up to 512 processors and scale to connect over 100,000 processors, which is particularly beneficial for AI deployments.
Broadcom's Tomahawk 6 chip can process up to 102.4 terabits per second, ideally suited for AI clusters needing substantial bandwidth for data-intensive workloads.
Cognitive Routing 2.0 enables the Tomahawk 6 to alleviate network bottlenecks by detecting congestion and redirecting traffic, while providing monitoring information.
Co-packaged optics (CPO) integrate transceivers directly into the switch processor, saving costs and reducing power consumption compared to traditional setups.
Tomahawk 6 supports clusters with up to 512 processors, facilitating networks with over 100,000 processors, enhancing scalability for large AI deployments.
Collection
[
|
...
]