The NVIDIA Rubin Platform: Six New Chips, One AI Supercomputer

2026-01-06 · Source: NVIDIA · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Cloud Computing & IT Infrastructure, Emerging Technologies & Innovation · Depth: Intermediate, quick

Summary

NVIDIA has unveiled the Vera Rubin AI supercomputer architecture, a system comprising six distinct chips engineered for unified operation. This architecture features the custom-designed Vera CPU, which doubles the performance of its predecessor, and the Reuben GPU, both co-designed for bidirectional, coherent, low-latency data sharing. The Vera Rubin compute board integrates 17,000 components, including one Vera CPU and two Reuben GPUs, delivering 100 petaflops of AI performance, a five-fold increase. Data transfer is handled by Connect X9, providing 1.6 terabits per second bandwidth per GPU, and the Bluefield 4 DPU, which offloads storage and security tasks. The system's compute tray is redesigned to eliminate cables, hoses, and fans, housing a Bluefield 4 DPU, eight Connect X9s, two Vera CPUs, and four Reuben GPUs. The MVL Link switch, now in its sixth generation, connects 18 compute nodes, scaling up to 72 Reuben GPUs, while Spectrum X Ethernet Photonix enables scaling to thousands of racks. The first Vera Rubin MVL 72 rack, the culmination of 15,000 engineer-years of development, contains six breakthrough chips, 18 compute trays, nine MVLink switch trays, and 220 trillion transistors, weighing nearly two tons.

Key takeaway

For CTOs and VPs of Engineering evaluating next-generation AI infrastructure, the Vera Rubin architecture represents a significant leap in compute density and interconnectivity. Your teams should consider its integrated CPU/GPU/DPU design and MVL Link capabilities for demanding AI workloads, especially those requiring massive scale-out and high data throughput. This system's design, emphasizing cable-free trays and advanced networking, could simplify deployment and maintenance for large-scale AI factories.

Key insights

The Vera Rubin architecture integrates co-designed CPUs, GPUs, and DPUs for unprecedented AI supercomputing performance and scalability.

Principles

• Co-design CPUs and GPUs for optimal data sharing.
• Offload non-compute tasks to DPUs.
• Eliminate cables for improved reliability and cooling.

In practice

• Utilize MVL Link for high-bandwidth GPU interconnects.
• Deploy Spectrum X Ethernet for large-scale AI factories.
• Integrate Bluefield DPUs for enhanced security and storage.

Topics

• Vera Rubin Architecture
• AI Processors
• High-Speed Interconnects
• Data Processing Units

Best for: CTO, VP of Engineering/Data, AI Architect, MLOps Engineer, Director of AI/ML

Related on AIssential

• Nvidia: Star Attraction at CES 2026 — Nvidia's Vera Rubin platform integrates co-designed chips and software to build scalable AI factories.
• The Rubin Era: How NVIDIA’s New Platform Rewrites the Rules for MoE and Agentic AI — NVIDIA's Vera Rubin platform redefines AI hardware by co-designing six chips for unprecedented performance in MoE and agentic AI.
• Inside the NVIDIA Rubin Platform: Six New Chips, One AI Supercomputer — NVIDIA's Rubin platform uses extreme co-design to optimize entire data centers for continuous, scalable AI intelligence production.
• How the NVIDIA Vera Rubin Platform is Solving Agentic AI’s Scale-Up Problem — Extreme co-design of hardware and software is crucial for economically scaling agentic AI workloads requiring low-latency and high-throughput.
• Nvidia Unveils Faster AI Chips Sooner Than Expected — Nvidia accelerates its AI chip release cycle to meet intense market demand and competition.
• B200 to Vera Rubin: What NVIDIA Changed Again — NVIDIA's Vera Rubin platform significantly boosts AI compute and memory bandwidth, introducing specialized hardware for long-context inference.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by NVIDIA.