NVIDIA Announces Next-Gen Rubin AI Chip

At COMPUTEX 2024, NVIDIA CEO Jensen Huang introduced the groundbreaking "Rubin" AI chip architecture, set to succeed the recently launched "Blackwell" architecture from March 2024. The Blackwell series, currently in production, is expected to begin shipping later this year. The Rubin architecture is scheduled for official release in 2026.

Huang highlighted NVIDIA's strategic shift to an accelerated release cycle, promising a new AI chip generation every year, a notable increase from the previous two-year cycle. This initiative underscores NVIDIA's commitment to maintaining its leadership position in the highly competitive AI chip market.

Looking ahead, NVIDIA plans to introduce the Blackwell Ultra in 2025, followed by the first generation of Rubin in 2026, and the Rubin Ultra in 2027. The Rubin architecture will feature support for 8-layer HBM4 high-bandwidth memory, with the Rubin Ultra extending this capability to 12 layers.

In addition, NVIDIA unveiled a new CPU, codenamed "Vera," which will be launched in tandem with the Rubin GPU to create the Vera Rubin superchip, poised to replace the current Grace Hopper platform.

The Rubin platform will also boast impressive features such as the next-generation NVLink 6 Switch with up to 3600 GB/s bandwidth and the CX9 SuperNIC component offering up to 1600 GB/s bandwidth.

These advancements represent NVIDIA's ongoing dedication to innovation and excellence in AI technology, ensuring they remain at the forefront of the industry.