Broadcom Launches High-Performance Fabric for AI Networks, Jericho3-AI


Broadcom has launched Jericho3-AI, a high-performance fabric for artificial intelligence (AI) networks. The new fabric is designed to drastically reduce the time it takes for AI workloads to complete a task. With “best-in-class” features like congestion-free operation, flawless load balancing, Zero-Impact Failover, and ultra-high radix, the Jericho3-AI is aimed at revolutionizing AI networking, enabling network administrators to manage the constantly growing workloads posed by AI demands.

The release of Jericho3-AI is timely as the usage of AI continues to surge. According to International Data Corporation (IDC), worldwide investment in AI will reach $154 billion in 2023 and $300 billion or more by 2026, an increase of 26.9% above the figure for 2022. Network administrators can rely on fabrics based on Jericho3-AI to handle the constantly growing AI workloads.

Ram Velaga, Senior Vice President and General Manager of Broadcom’s Core Switching Group, said that reduced training and inference times for large-scale AI models are the norm for AI networking. “When compared to other market alternatives, Jericho3-AI significantly reduces job completion time.” When an AI compute cycle is completed, a minimal number of big, long-lived flows begin simultaneously, which is one of the distinctive properties of AI workloads.”

Jericho3-AI fabric would offer the best performance for certain workloads with special functionality created specifically for AI applications, such as optimal load balancing, end-to-end traffic scheduling, congestion-free operation, and zero-impact failover capability. With the Jericho3-AI fabric’s ability to grow connections to 32,000 GPUs, each with 800Gbps, in a single cluster, it would offer ultra-high radix. By using this exclusive capability, Jericho3-AI fabric would offer the fastest task completion times.

AI Networking

According to Broadcom, the Jericho3-AI fabric would deliver 40% less power per gigabit while providing 26 petabits per second of Ethernet capacity, almost four times as much as the previous generation.

Bob Wheeler, Chief Analyst at Wheeler’s Network, said “cloud operators will upgrade their AI infrastructure to address the massive growth in bandwidth, driven by a new generation of high-capacity GPUs and the emergence of large language models. Jericho3-AI offers a high-bandwidth, low-latency, and low-power choice for networks connecting tens of thousands of GPUs, revolutionizing the economics of building and maintaining AI clusters for this exciting new era.”

Jericho3-AI would stand out in the market by delivering maximum performance while also allowing for the lowest total cost of ownership (TCO), made possible by features like distributed buffering, long-reach SerDes, and enhanced telemetry, all of which are provided via industry-standard Ethernet. These features would provide the broadest ecosystem of hardware and software vendors, as well as a great degree of flexibility in network topologies and deployment options, according to Broadcom.

Broadcom anticipates the launch of Jericho3-AI is a significant milestone in the AI networking space, as it provides the highest performance fabric for AI networks. Customers that meet the requirements may now purchase Jericho3-AI.

“The Arista 7800R3 running EOS is the premier AI spine supporting exponential growth of AI workloads and network traffic at petabit scale,” said Anshul Sadana, Chief Operating Officer & Senior Vice President, Arista. “We are excited to collaborate with Broadcom and drive the AI revolution using the innovative Ethernet-based cluster approach demonstrating groundbreaking performance in reducing the runtime of AI training and inference workloads.”