The Nvidia Rubin is a next-generation AI superchip, featuring both a Rubin GPU and a Vera CPU. It's designed to succeed the Blackwell architecture and will utilize HBM4 memory.
- the Rubin GPU will be paired with the Vera CPU, a custom-designed processor, to work together seamlessly.
- mass production is slated for late 2025, with availability expected in early 2026.
- it will feature advanced HBM4 memory and is expected to deliver significant performance improvements, particularly in AI training and inference tasks.
- Vera CPU: The Vera CPU is Nvidia's first custom-designed CPU and is built to work in close coordination with the Rubin GPU.
- release timeline: The Vera Rubin superchip is scheduled for release in 2026, with initial availability expected in the second half of the year.
- key technologies: The Rubin architecture will incorporate TSMC's 3nm process and leverage NVIDIA's first-ever chiplet design. It also includes a new NVLink 6 architecture and next-generation networking components like CX9 smart NICs.
- performance Ggoals: The Vera Rubin system aims to achieve a significant increase in performance compared to previous generations, potentially multiplying exaflops by 15 and scaling the future of AI infrastructure. For example, a fully equipped NVL 144 rack system is expected to deliver 3.6 exaflops of FP4 inference compute.
