Cables and wires have been completely eliminated from Nvidia’s newest AI computing hardware in favor of total liquid cooling. Jensen Huang revealed a Vera Rubin node on stage, announcing that the system has entered full production and takes less than 5 minutes to assemble, compared to the 2 hours required for Grace Blackwell. Each node incorporates 2 GPUs and employs 100% liquid cooling.
The new architecture replaces the 80% liquid cooling used in the Grace Blackwell design. Huang stated that the simplified construction translates to higher manufacturing cycle times, as well as increased reliability and better performance compared to previous iterations.
