NVIDIA unveils AI accelerator upgrade plans, with Blackwell Ultra coming in 2025 and Rubin debuting in 2026
NVIDIA has announced plans to upgrade its AI accelerators every year, revealing that it is developing Blackwell Ultra chips for 2025 and a next-generation platform called Rubin for 2026.
'Accelerate Everything,' NVIDIA CEO Says Ahead of COMPUTEX | NVIDIA Blog
NVIDIA Confirms Next-Generation Architecture is Based on (Vera) Rubin GPU
Jensen Huang Computex Keynote: Nvidia Reveals New AI Software and Services - Bloomberg
https://www.bloomberg.com/news/articles/2024-06-02/jensen-huang-computex-keynote-nvidia-reveals-new-ai-software-and-services
NVIDIA Unveils Next-Gen Rubin, Rubin Ultra, Blackwell Ultra GPUs & Supercharged Vera CPUs
https://wccftech.com/nvidia-unveils-next-gen-rubin-rubin-ultra-blackwell-ultra-gpus-supercharged-vera-cpus/
NVIDIA CEO Jensen Huang, who delivered a keynote speech at the eve of COMPUTEX 2024, a computer trade fair held in Taipei, said he sees the recent rise of generative AI as a new industrial revolution, and expects more companies and institutions to adopt AI beyond the handful of cloud computing companies that have contributed greatly to his company's sales growth.
With these expectations as a backdrop, NVIDIA has demonstrated its commitment to continuing to invest in AI products and has also revealed its upcoming product lineup.
First, Blackwell Ultra , an extended version of the GPU architecture Blackwell, which was announced in March 2024, is scheduled to be released in 2025. Blackwell Ultra is a model that adds 12-Hi HBM3e compatibility to Blackwell, and has improved performance over Blackwell.
NVIDIA announces GPU architecture 'Blackwell' and new GPU 'B200' to realize AI models with trillions of parameters - GIGAZINE
Next, Rubin , the successor to Blackwell, was unveiled. Rubin supports 8-Hi HBM4 and features integration with NVIDIA hardware innovations such as 'new GPUs, new Arm-based CPU 'Vera', advanced networking with NVLink 6, CX9 SuperNIC, and X1600 converged InfiniBand/Ethernet switches.' NVLink 6 is expected to provide data transfer speeds of up to 3600GB/s, and CX9 SuperNIC is expected to provide network speeds of up to 1600Gb/s. The initial lineup with Rubin will include the NVIDIA Rubin R100 GPU, which is expected to begin mass production in the fourth quarter of 2025. Meanwhile, systems such as DGX and HGX solutions are scheduled to be mass produced in the first half of 2026.
In addition, 'Rubin Ultra', which is compatible with 12-Hi HBM4, will be released in 2027.
'We are witnessing computational inflation. As the amount of data to be processed grows exponentially, traditional computing methods can no longer keep up, and only NVIDIA's accelerated computing can reduce costs,' Huang said.
Apart from these architectures, NVIDIA also announced new designs for server computers using its chips.
Related Posts:
in Hardware, Posted by log1p_kr