Nvidia Introduces Cableless Computing Tray for AI

Nvidia has showcased its new Vera Rubin computing tray, engineered to significantly reduce cable usage in the assembly of AI data centers. This new design aims to accelerate and simplify the installation and maintenance of thousands of GPU units.

The company states that the new Vera Rubin tray, housing Nvidia's latest chips, has achieved a near cableless configuration. This is facilitated by liquid cooling, which replaces the numerous cables required for power and network connections with a single, centralized connector. Nvidia refers to this as 'cableless,' though two cables remain necessary for power and networking.

Andrew Bell, Nvidia's Senior VP of Hardware Engineering, noted that older models featured up to 50–60 cables, taking two to three hours to assemble with only a 20% success rate. He claims the new Vera Rubin tray can be assembled in five minutes with a 95% success rate.

Reducing cables not only speeds up assembly and minimizes errors but also improves component airflow and simplifies servicing. The liquid cooling approach enables higher performance and lower power consumption compared to traditional air-cooling methods.

The development of Nvidia's Vera Rubin platform reflects a broader industry trend towards optimizing data centers and increasing automation. By reducing reliance on cables and enhancing cooling efficiency, Nvidia aims to meet the growing demand for artificial intelligence computing power.