The NVIDIA Vera Rubin platform, a full-stack AI supercomputing platform for agentic AI, was announced at GTC 2026. Basic documentation for this platform can primarily be found through official NVIDIA channels, specifically their corporate website, technical blogs, and dedicated resource hubs. Given the recency of its announcement, detailed developer documentation might still be in the process of being fully rolled out, but several key resources are available.
NVIDIA’s main website and their GTC 2026 announcements are crucial starting points. The NVIDIA Vera Rubin platform integrates several components, including the NVIDIA Vera CPU, NVIDIA Rubin GPU, NVIDIA NVLink™ 6 Switch, NVIDIA ConnectX®-9 SuperNIC, NVIDIA BlueField®-4 DPU, and NVIDIA Spectrum™-6 Ethernet switch, along with the newly integrated NVIDIA Groq 3 LPU. Information regarding these individual components and how they coalesce within the Vera Rubin architecture is detailed in press releases and overview pages. For instance, the NVIDIA Vera Rubin POD page provides an overview of the system’s architecture, highlighting its seven chips and five rack-scale systems. Furthermore, NVIDIA has released the “NVIDIA Vera Rubin DSX AI Factory reference design,” which offers guidance on designing, building, and operating the entire AI factory infrastructure stack, covering compute, networking, and storage. This reference design includes documentation on best practices for power, cooling, and control systems, ensuring seamless hardware-software integration and scalable deployment.
For more in-depth technical specifications and architectural breakdowns, developers should consult NVIDIA’s dedicated resource hubs and technical blogs. These platforms typically host whitepapers, developer guides, and detailed articles that delve into the specifics of NVIDIA’s hardware and software offerings. For example, some resources discuss the Vera CPU, a custom ARM processor designed for agentic AI workloads, and the Rubin GPU, which features HBM4 memory and a new Transformer Engine. The platform also emphasizes its focus on agentic AI and reasoning, with capabilities designed to handle multi-step problem-solving and massive long-context workflows. Developers can also find information on the NVIDIA Omniverse DSX Blueprint, which is compatible with the Vera Rubin DSX AI Factory reference design and enables the creation of physically accurate digital twins for AI factory design and simulation. While specific, granular programming guides for the Vera Rubin platform might evolve over time, the foundational documentation on NVIDIA’s official sites provides a comprehensive understanding of its architecture and intended use cases for building large-scale AI infrastructure.