Arista Networks, a global vendor delivering cloud and AI networking solutions (opens PDF in new window), has unveiled the Arista Etherlink AI platforms, aimed at enhancing network performance for demanding AI workloads such as training and inferencing. These platforms are powered by the new AI-optimized Arista EOS features, supporting AI cluster sizes from thousands to hundreds of thousands of XPUs.

These Arista AI networking platforms would utilize efficient one and two-tier network topologies for superior application performance and advanced monitoring capabilities.

Alan Weckel, Founder and Technology Analyst for 650 Group, emphasized the importance of network performance in AI clusters, noting that the Arista Etherlink AI platforms provide an 800G end-to-end technology platform across various network segments, benefiting from well-established Ethernet tools and security.

The Arista Etherlink AI portfolio includes the 7060X6 AI Leaf switch, featuring Broadcom Tomahawk 5 silicon with a capacity of 51.2 Tbps, and the 7800R4 AI Spine, incorporating the latest Broadcom Jericho3-AI processors for non-blocking throughput. The 7700R4 AI Distributed Etherlink Switch (DES) supports the largest AI clusters with massively parallel distributed scheduling and congestion-free traffic spraying.

These platforms can support over 10,000 XPUs with a single-tier network topology and more than 100,000 XPUs with a two-tier network, optimizing AI application performance, reducing optical transceiver usage, lowering costs, and improving reliability. All Etherlink switches are compatible with emerging Ultra Ethernet Consortium (UEC) standards, expected to further enhance performance with UEC NICs.

Ethernet for AI Workloads

Ram Velaga, Senior Vice President and General Manager of the Core Switching Group at Broadcom, praised the versatility and robustness of Ethernet for AI workloads, highlighting the superiority of Ethernet-based solutions like those from Arista over proprietary technologies.

Arista’s EOS and CloudVision software suites would complement the new hardware, providing AI-for-networking, security, segmentation, visibility, and telemetry features. Innovations like RDMA-aware load balancing, QoS, and AI Analyzer enhance configuration, visibility, and performance analysis for AI workloads.

Hugh Holbrook, Chief Development Officer at Arista Networks, underscored the company’s competitive advantage stemming from its rich operating system and broad product portfolio, designed to meet the needs of AI networks of all sizes. The AI-optimized EOS features enable faster deployment, reduced configuration issues, and improved AI job completion times.

The Arista 7060X6 is already available, while the company’s 7800R4-AI and 7700R4 DES are undergoing customer testing and are expected to be available in the second half of 2024.