NVIDIA Unveils Next-Gen AI: Rubin, Six New Chips, a Powerful Supercomputer

NVIDIA has officially introduced its next-generation AI platform, the Rubin, during CES 2023. This powerful architecture includes six new chips designed to advance AI technology significantly. The Rubin platform, named after pioneering astronomer Vera Rubin, aims to break new ground in building, deploying, and securing extensive AI systems at reduced costs.

NVIDIA Rubin Platform Overview

The Rubin platform leverages extreme codesign across its hardware and software components. This approach allows for a substantial 10-fold reduction in inference token costs. Additionally, it enables training models using four times fewer GPUs compared to NVIDIA’s previous Blackwell platform.

Key Components of the Rubin Architecture

• NVIDIA Vera CPU: Optimized for large-scale AI applications, featuring 88 custom cores and exceptional bandwidth capabilities.
• NVIDIA Rubin GPU: Delivers 50 petaflops of compute power for AI inference with innovative adaptive compression.
• NVIDIA NVLink 6 Switch: Facilitates high-speed GPU-to-GPU communication, essential for large mixture-of-experts (MoE) models.
• NVIDIA ConnectX-9 SuperNIC: Enhances network performance, ensuring efficient data processing.
• NVIDIA BlueField-4 DPU: Powers the new Inference Context Memory Storage Platform for better AI reasoning capabilities.
• NVIDIA Spectrum-6 Ethernet Switch: Increases overall power efficiency and reliability in AI networking.

Advancements in AI Computing

Jensen Huang, CEO of NVIDIA, highlighted the urgent demand for enhanced AI computing capabilities. He remarked, “Rubin arrives at exactly the right moment, as AI computing demand for both training and inference is going through the roof.” The new architecture promises to reshape AI infrastructure through advanced processing and reasoning functionalities.

Collaboration with Major Corporations

Several leading companies have expressed their commitment to integrating the Rubin platform into their operations:

• Amazon Web Services (AWS): Plans to deploy Rubin-based instances to enhance AI services.
• Microsoft: Will utilize NVIDIA Vera Rubin in next-gen AI superfactories and cloud services.
• Google: Continues its partnership with NVIDIA to provide optimal environments for AI applications.
• CoreWeave: Aims to incorporate Rubin systems to facilitate advanced AI workloads.

Future Availability

The Rubin platform is set for full production, with products expected from major cloud providers in the second half of 2026. This initiative will significantly accelerate AI adoption across sectors, supported by robust partnerships with organizations like Red Hat, Oracle, and various hardware manufacturers.

Conclusion

NVIDIA’s Rubin platform stands to redefine AI infrastructure with its innovative technologies and collaborative ecosystem. The introduction of Rubin marks a pivotal moment in AI development, promising to deliver unprecedented efficiency and performance for diverse AI workloads.