Nvidia has entered into full production of its hyperscale platform, Vera Rubin, the third generation of the company’s rack-scale systems built for AI data centers.
The company said that Taiwan server vendors as well as other supply chain operators are manufacturing Vera Rubin-based systems at scale designed for AI labs, cloud farms and hyperscalers.
Vera Rubin is five purpose-built racks operating as one massive AI supercomputer targeted at agentic workloads. The fully integrated system includes:
- Vera Rubin NVL72 systems
- Nvidia Vera CPU
- Nvidia Groq 3 LPX
- Nvidia Vera BlueField-4 STX storage
- Nvidia Spectrum-6 SPX Ethernet racks
The Spectrum-X Ethernet Photonics is billed as the world’s first co-packaged-optics (CPO)-based switches with 200 Gb/s SerDes, a technology that is also now in production, Nvidia said.
Spectrum-X Ethernet Photonics delivers five times better power efficiency, five times longer AI uptime and 1.3 times faster time to deployment than networks using traditional transceivers, Nvidia said.
10 times agent throughput
The system delivers 10 times agent throughput at scale compared to Nvidia’s previous generation called Grace Blackwell, the company claims.
“Agentic AI is a new kind of workload,” said Jensen Huang, founder and CEO of Nvidia. “One prompt can launch a thousand-step journey of reasoning, retrieval, tool use and response generation.”
Vera Rubin is the third generation of data center rack-scale systems from Nvidia.
Those companies building systems with Vera Rubin includes Dell Technologies, HPE, Lenovo and Supermicro, as well as AIC, Aivres, ASRock Rack, ASUS, Cloudian, Compal, DDN, Everpure, Foxconn, GIGABYTE, Hitachi Vantara, Hyve Solutions, IBM, Inventec, MinIO, MiTAC Computing, MSI, NetApp, Nutanix, Pegatron, Quanta Cloud Technology (QCT), VAST Data, WEKA, Wistron and Wiwynn.
