Tying a number of knowledge facilities collectively into an built-in entire is a sizzling matter in knowledge middle circles. The objective is to attain efficiency past what a single knowledge middle or AI manufacturing unit can ship. However with energy calls for rising, integrating knowledge facilities throughout a number of location the place there’s obtainable energy is smart.
However combining knowledge facilities into an enormous mixture whereas overcoming the networking limitations of off-the-shelf Ethernet — issues akin to jitter, excessive latency and erratic efficiency — is the problem.
Forward of subsequent week’s Scorching Chips 2025 convention in Palo Alto, NVIDIA right now introduced Spectrum-XGS Ethernet, which the corporate calls a scale-across expertise for combining a number of, distributed knowledge facilities into “unified, giga-scale AI super-factories.”
NVIDIA mentioned the Spectrum-X Ethernet networking platform gives 1.6x better bandwidth density than off-the-shelf Ethernet for multi-tenant, hyperscale AI factories. It includes NVIDIA Spectrum-X switches and NVIDIA ConnectX-8 SuperNICs, designed to ship scalability and low latency.
“The AI industrial revolution is right here and large scale AI factories are the important infrastructure,” mentioned Jensen Huang, founder and CEO of NVIDIA. “With NVIDIA Spectrum-XGS Ethernet, we add scale-across to scale-up and scale-out capabilities to hyperlink knowledge facilities throughout cities, nations and continents into huge, giga-scale AI super-factories.”
The corporate mentioned Spectrum-XGS Ethernet gives auto-adjusted distance congestion management, latency administration and end-to-end telemetry. The providing almost doubles the efficiency of the NVIDIA Collective Communications Library, in keeping with NVIDIA, accelerating multi-GPU and multi-node communication to ship predictable efficiency throughout geographically distributed AI clusters. In consequence, a number of knowledge facilities can function as a single AI super-factory, totally optimized for long-distance connectivity.
The brand new providing additionally builds on NVIDIA’s partnership with AI hyperscaler CoreWeave, which NVIDIA mentioned will likely be among the many first to attach its knowledge facilities with Spectrum-XGS Ethernet.
“CoreWeave’s mission is to ship essentially the most highly effective AI infrastructure to innovators all over the place,” mentioned Peter Salanki, co-founder and CTO of CoreWeave. “With NVIDIA Spectrum-XGS, we will join our knowledge facilities right into a single, unified supercomputer, giving our clients entry to giga-scale AI that may speed up breakthroughs throughout each trade.”
