FS has announced its PicOS AI Switch System, a data center networking solution designed for large-scale artificial intelligence (AI) training, inference, and high-performance computing (HPC) workloads. The system integrates Broadcom Tomahawk series switching chips, FS’s PicOS network operating system, and the AmpCon-DC management platform.
According to FS, the PicOS AI Switch System is engineered for lossless RDMA over Converged Ethernet version two (RoCEv2) networking, ultra-low latency, and intelligent traffic optimization. The solution is designed to maximize GPU efficiency and deliver reliable cluster performance in high-density data center environments.
The PicOS AI Switch portfolio includes 400 gigabit (400G) and 800 gigabit (800G) models, featuring Broadcom Tomahawk 3, 4, and 5 series chips. FS says these switches provide high bandwidth, deterministic performance, and scalable connectivity, meeting the demands of AI training, inference, and HPC deployments.
FS highlights features such as redundant architecture, deep buffers, and advanced congestion management to ensure lossless performance and operational resilience. The AmpCon-DC platform enables deployment, configuration, and lifecycle management of large GPU-based networks, supporting faster scaling and reducing operational complexity for data center operators.
The main application for this system, as explicitly stated by FS, is in building scalable, high-performance GPU clusters for data centers involved in AI and accelerated computing projects. Additional applications noted by FS outside the core data center market include the broader enterprise, telecom, and HPC sectors.
Source: FS
