Chip giant Broadcom introduces Jericho series for enhancing AI-driven data center networks
Broadcom's Jericho4: Revolutionizing AI-Driven Data Center Networking
Broadcom has unveiled its next-generation Ethernet routing chip, the Jericho4, designed to cater to the growing needs of AI-driven data centers and large-scale distributed AI infrastructures.
The Jericho4 boasts a deep-buffered Ethernet throughput of 51.2 Tbps, enabling lossless, low-latency networking optimized for demanding AI workloads. This ensures efficient scaling across racks, clusters, and geographically distributed data centers without performance degradation.
One of the key features of Jericho4 is its support for distributed AI workloads that exceed the capacity of single data centers. It enables the interconnection of over a million XPUs (accelerators like GPUs and TPUs) across multiple facilities, addressing challenges related to transmitting lossless, high-bandwidth, and low-latency data over metropolitan and regional distances.
The chip also features 3.2 Tbps HyperPort interfaces for very high throughput connections, enhancing its ability to handle large volumes of network traffic efficiently. Advanced congestion control and support for long-distance RDMA over Converged Ethernet (RoCE) transport further improve data transfer reliability and speed essential for distributed AI computations.
To ensure secure data transport over long distances, Jericho4 incorporates MACsec encryption, providing data integrity and confidentiality in multi-data-center environments. The chip is manufactured using the 3nm semiconductor process, offering lower power consumption and improved system reliability, supporting energy efficiency in large-scale AI and data center operations.
The Jericho4's deep buffering technology helps to absorb traffic bursts and prevents packet loss during network congestion, crucial for maintaining AI workload performance. This, coupled with its high-bandwidth memory similar to that used by Nvidia and AMD for their AI processors, makes it an ideal solution for speeding artificial intelligence computation.
In essence, Jericho4 combines very high bandwidth, lossless and low-latency transport, deep buffering, advanced security, and energy-efficient design to meet the growing needs of AI-scale Ethernet fabric deployments spanning multiple data centers and regions. This positions it as a leading solution for networking in AI infrastructure at massive scale.
By introducing and improving features that increase the amount of networking traffic speeding across large networks, the Jericho4 chip is aimed at improving the efficiency of data transfer in data centers, particularly for cloud computing companies. Its ability to connect data centers over 60 miles (96.5 km) apart further underscores its potential to revolutionize AI-driven data center networking.
- Technology like the Jericho4, a data-and-cloud-computing innovation from Broadcom, is designed to optimize networking for demanding AI workloads, thereby enhancing the efficiency of data transfer in data centers, especially for cloud computing companies.
- The Jericho4 chip's remarkable capabilities, including its support for distributing AI workloads across multiple facilities and connecting data centers over long distances, suggest that it could revolutionize data-and-cloud-computing technology, particularly in the realm of AI-driven data center networking.
