Nvidia Unveils Cableless AI Trays With Vera Rubin Platform

Nvidia revealed its new "cableless" AI trays during a media event at an "engineering superlab" in Silicon Valley. These trays, designed to slide into racks within data centers, represent a significant redesign aimed at improving efficiency and simplifying integration. The new design, exemplified by the Vera Rubin platform, drastically reduces the number of visible cables compared to previous generations.

Andrew Bell, Senior VP of Hardware Engineering at Nvidia, explained that the trays are "cableless" rather than literally cable-free, with two essential cables remaining: one for power and another for networking. These cables are designed to be unobtrusive, a stark contrast to the "rat's nest" of cables found in older tray designs. The Vera Rubin tray was displayed alongside a previous-generation tray based on Nvidia's GB200 and GB300 chips, highlighting the visual and functional improvements.

During the visit, Nvidia demonstrated real AI tasks being processed on its new platform. Among these tasks was the testing of a version of OpenAI's GPT model, optimized for the Vera Rubin platform. This collaboration underscores the advanced capabilities and potential applications of Nvidia's latest hardware in the rapidly evolving field of artificial intelligence. The "engineering superlab" itself, an unmarked facility, houses hundreds of racked GPUs actively performing AI computations, illustrating the scale of Nvidia's operations.