At the Computex 2024 keynote, NVIDIA CEO Jensen Huang reiterated the company’s commitment to driving technological growth with a “One Year Rhythm” approach. He also previewed the upcoming release of GPUs based on the Blackwell architecture, with the Blackwell Ultra GPU set for launch in 2025. Furthermore, Huang confirmed that the next generation of graphics architecture, codenamed “Rubin,” will include the Rubin Ultra GPU, slated for release in 2026.
However, NVIDIA has yet to disclose specific details about the Blackwell Ultra GPU and the Rubin GPU products. It was revealed that the Blackwell Ultra GPU will feature a 12-stack design with 8S HBM3e 12H high-density memory modules, while the Rubin GPU will utilize 8S HBM4 high-density memory modules, and the Rubin Ultra GPU will adopt 12S HBM4 high-density memory modules. These configurations are intended to provide higher data transfer bandwidth and memory capacity, thereby meeting the demands of advanced AI computational data throughput.
In addition to the yearly release of new GPU products, NVIDIA also confirmed the forthcoming launch of a CPU product codenamed “Vera,” along with the sixth-generation NVLink design and the Spectrum-X1600 Ethernet platform. These developments aim to accelerate the growth of large-scale AI applications.
The Rubin GPU is named after the pioneering American astronomer Vera Rubin, known for her research on galaxy rotation rates. Accordingly, the upcoming “Vera” CPU product also derives its name from her.
1 thought on “Jensen Huang Confirms Blackwell Ultra GPU for 2025, Rubin Ultra for 2026”