FS introduced its new 400G and 800G Ethernet switches aimed at large AI clusters, leveraging the Broadcom Tomahawk 5 chipset. The switches deliver up to 51.2 Tbps (102.4 Tbps Full Duplex) of switching capacity, with support for lossless RoCEv2 networking, providing the low-latency performance critical for AI, big data, and high-performance computing (HPC) applications.
The rapid growth of generative AI has increased the bandwidth demands for AI/ML clusters. FS responded with these high-performance 400G and 800G switches, optimized for efficient data transmission with reduced latency. These models support a range of configurations, including breakout cables for multiple Ethernet speeds, further enhancing flexibility in networking architectures.
The 400G switch comes with 128x QSFP112 ports, scaling up to 32K 400G ports in Layer 3 configurations. The 800G model offers 64x OSFP ports, capable of breaking out into 8x 100GbE connections. The switches are designed to improve bandwidth efficiency with advanced load-balancing features, ensuring consistent performance even under heavy AI workloads.
Key Features of FS 400G and 800G AI Switches:
• Powered by Broadcom Tomahawk 5, offering 51.2 Tbps (102.4 Tbps Full Duplex) of switching capacity.
• Lossless RoCEv2 networking for low-latency, zero-packet-loss performance.
• High-density port configurations for 100/200/400/800G in a 4 RU chassis.
• DLB and GLB load balancing technologies for optimized bandwidth usage.
• Adaptive routing for congestion control based on real-time network conditions.
Daniel Xiang, CEO of FS, said, “With the launch of our 400G and 800G AI switches, enterprises can leverage greater network capacity and efficiency for their AI clusters. Based on Broadcom’s TH5 chips, these switches deliver high bandwidth and throughput needed for AI workloads.”
