NVIDIA has announced its latest networking technology, Spectrum-XGS Ethernet, a groundbreaking platform designed to connect geographically dispersed data centers into a single, massive AI super-factory. This new “scale-across” technology aims to overcome the physical and power limitations of individual data centers, enabling the creation of unprecedented AI computing power.
As the demand for AI continues to surge, individual data centers are hitting their limits in terms of power and capacity. To address this, NVIDIA’s Spectrum-XGS Ethernet offers a solution to seamlessly connect multiple data centers, effectively creating a unified infrastructure. This approach tackles the challenges of high latency and unpredictable performance that have traditionally hindered the use of standard Ethernet over long distances.
The Spectrum-XGS platform incorporates advanced congestion control, latency management, and telemetry, which can nearly double the performance of the NVIDIA Collective Communications Library. This allows for efficient and high-speed communication between distributed data centers, making them function as one cohesive AI super-factory.
Cloud provider CoreWeave is among the first to adopt this innovative technology, planning to connect its data centers with Spectrum-XGS. This move is expected to significantly enhance their AI and high-performance computing capabilities.
The Spectrum-XGS platform is part of the broader Spectrum-X Ethernet platform, which also includes Spectrum-X switches and ConnectX-8 SuperNICs. NVIDIA has also recently introduced other networking innovations, such as the Quantum-X silicon photonics networking switches, further solidifying its position as a leader in high-performance networking for AI.
