Broadcom first to add AI to network switch chip
Broadcom has added a neural-network inference engine to its latest networking switch that can improve the performance of data networks used for AI applications.
The NetGNT (Networking General-purpose Neural-network Traffic-analyzer) is included in the software-programmable Trident 5-X12 chip and works in parallel to augment the standard packet-processing pipeline and can be trained to look for different types of traffic patterns across the entire chip.
The standard pipeline is one-packet/one-path, meaning that it looks at one packet as it takes a specific path through the chip’s ports and buffers. NetGNT, in contrast, can look for different types of traffic patterns in AI/ML workloads called “incast” in which many packet flows all converge on the same port and buffer at roughly the same time causing congestion. NetGNT can recognize this pattern in real-time as it begins and invoke congestion-control techniques to avoid degraded network performance.
As the NetGNT runs in hardware at full line rate there is no impact to throughput or latency and enhanced telemetry on the chip allows deeper real-time insights into network operations, which can then be used to train neural engine.
- 4Tbit/s Ethernet switch doubles bandwidth, cuts power
- World’s first 51.2 Tbit/s Ethernet switch chip plans chiplet
“We continue to push the envelope and introduce brand-new technologies such as NetGNT to the market. We also listen very closely to our customers who have made it clear that one size does not fit all. They rely on us to deliver a broad portfolio of chips, customized for different applications,” said Ram Velaga, senior vice president and general manager, Core Switching Group, Broadcom. “Trident 5-X12 is the most power-efficient ToR (Top of Rack) on the market, while still adding cutting-edge new features that our customers have come to expect from an innovator such as Broadcom.”
Trident 5-X12 is software-programmable, field-upgradable and provides 16.0 Tbit/s of bandwidth, which is double that of the market-leading Trident 4-X9 and uses 25% less power per 400G port.
The chip also adds support for 800G ports using 100G-PAM4 SerDes, allowing direct connection to Broadcom’s leading-edge Tomahawk 5, which is used as the spine/fabric in the newest compute and AI/ML data centres. This chip enables a 1RU data centre top of rack (ToR) switch supporting 48x200G downlink ports and 8x800G uplink ports.
“Broadcom’s Trident product family has long been an integral part of our Predictable Network Infrastructure, the foundation of reliable and secure service worldwide. We are strongly encouraged by the new Trident 5 introduction with increased performance, density and newly introduced NetGNT AI Engine. Trident family’s Networking Programing Language in-pipeline programmability helps us leverage existing investment, while quickly delivering new robust and innovative services,” said Dennis Cai, Head of Global Network Infrastructure, Alibaba Cloud.
The Trident 5-X12 is fully programmable using NPL (Network Programming Language), allowing in-field upgrades to add features such as new telemetry, security and traffic engineering
Broadcom is now shipping Trident 5-X12 devices to qualified customers.
If you enjoyed this article, you will like the following ones: don't miss them by subscribing to :
