NVIDIA Rubin Platform confirmed by Jensen during Computex 2024 keynote

Vera Rubin will be the patron of NVIDIA’s next GPU architecture.

NVIDIA made an announcement at Computex 2024 regarding its new GPU architecture. What initially seemed like a rather uneventful keynote on artificial intelligence took a turn when NVIDIA unveiled updates on its data-center architecture. The company confirmed that the next-generation GPU architecture would be named Rubin and would be paired with the Vera CPU.

Vera Rubin was chosen as the namesake for NVIDIA’s new architecture for data centers, which will utilize the newer HBM4 memory. NVIDIA has officially announced that the Rubin GPU will feature 8 stacks of HBM4 memory, with plans for a later release of the Rubin Ultra featuring 12 HBM4 stacks.

Furthermore, NVIDIA plans to introduce a successor to the Grace Hopper and Grace Blackwell superchips in the form of a new Vera Rubin board featuring the Vera CPU. While NVIDIA has not disclosed the specifications or architecture of this processor, the platform will also support brand-new NVLink 6 switches offering speeds of up to 3600 GB/s, along with the CX9 SuperNIC delivering speeds of up to 1600 Gb/s.

NVIDIA Data-center Roadmap, Source: NVIDIA/ HardwareLuxx

NVIDIA also announced the Blackwell Ultra ; however, this appears to be nothing more than the already announced GB200 GPU processor.

The Rubin architecture does not have a release date, and Jensen did not confirm when such GPUs should be expected. Earlier rumors suggested that the R100 processor would be released between Q4 2025 and Q1 2026.

Source: NVIDIA, HardwareLuxx