Spectrum-X has been fully produced, five times more energy efficient than the traditional network

News from IT House, June 2, announced on May 31 that its next generation super-calculating platform for the intelligent AI factory, NVIDIA Vera Rubin, was in full production. The IT House has been reported earlier.

In addition to this, we are also confirming that a new generation of Spectrum-X-Ethernet silicon-ray technologies have simultaneously entered full-scale production, which is the central building block of the platform ' s ability to connect large-scale AI plant networks.

As a global first-rate Ethernet switch, based on PVC (CPO), Spectrum-X uses 200Gb/s SerDes to integrate photocom components directly into the chip envelope.

This new technology, which is designed in conjunction with the use of a traditional luminous transceiver, will result in a five-fold increase in energy efficiency, a five-fold increase in AI ' s overall normal operation time, and a 1.3-fold acceleration in deployment time, providing a basic network structure for the construction of a million-grade GPU AI plant. CoreWeave, Lambda and Oracle Cloud Insurance were the first ecosystem partners to adopt the technology.

The production of the NVIDIA Vera Rubin platform marks the official launch of the NVIDIA ' s 3rd-generation flagship AI structure into the commercial delivery cycle, following Hopper and Blackwell.

Vera Rubin is the largest computational cluster-level platform in the history of NVIDIA, which combines five specially designed cabinets into a large AI supercomputer working together.

The platform is a full-fledged end-to-end solution by integrating the NVIDIA Vera Rubin NVL72 system, Vera CPU, Groq 3 LPX, Bluefield-4 STX storage unit and Spectrum-6 SPX Ether.

According to NVIDIA founder and CEO Hoang In-hoon, Vera Rubin is an AI plant engine for the new load of intelligent body AI, which has the high performance, efficiency and security required for large-scale output intelligence to lead the next industrial revolution.

In terms of performance, the Vera Rubin NVL72 cabinet-level configuration can achieve a maximum of 260 TB/s total interconnective bandwidth through the sixth generation of NVLink exchange technology and provide full interconnectivity bandwidth up to 3.6 TB/s per GPU.

Vera Rubin achieved 10 times more smart body throughput on a scale than the last generation of Grace Blackwell platforms. The platform has also experienced a rapid expansion at the supply chain level, with production at twice the scale of the former Grace Blackwell, and several hundred supply chain partners in over 350 factories around the world are accelerating the upgrading of capacity.

In order to accelerate the deployment of the AI plant, NVIDIA simultaneously launched the DSX platform for the Vera Rubin POD architecture, providing the full design and operational basis for the AI plant. The DSX programme has been fully implemented for the accelerated deployment of the Vera Rubin AI plant, including Dell, Hye Cai, Consort, Super Four Head Vendors, as well as core industries such as Hua-seok, Fuscon, Technological, and Suk-seok, Guangda, and Latitue.

The Vera Rubin platform also deeply integrates Bluefield-4 DPU, providing software definition networks and hardware-level multi-tensor segregation capabilities. At the same time, the platform provides a credible implementation environment for the AI plant at the shelf level through the full house NVIDIA classified computing technology. Vera Rubin's official shipment is expected to start in autumn 2026.

Read about:

Advertising statements: The external jump links (including not limited to hyperlinks, 2D codes, passwords, etc.) contained in the text are used to convey more information and save time for selection purposes only for reference purposes, which are included in all IT House articles。