The NVIDIA Vera Rubin platform, designed as a full-stack AI supercomputing platform for agentic AI, integrates a sophisticated array of software layers to manage, optimize, and secure complex AI workflows. These software components work in conjunction with the platform’s advanced hardware, which includes Vera CPUs, Rubin GPUs, Groq 3 LPUs, and various networking and DPU technologies, to create a cohesive and highly efficient AI factory. The software stack focuses on enabling large-scale pretraining, post-training, and real-time agentic inference, addressing the unique demands of autonomous AI systems.
A critical component of Vera Rubin’s software ecosystem is the NVIDIA DSX Platform, encompassing DSX Max-Q and DSX Flex. DSX Max-Q is designed for dynamic power provisioning across the entire AI factory, allowing for the deployment of more AI infrastructure within a fixed power data center by optimizing power usage. Complementing this, DSX Flex software enables AI factories to function as grid-flexible assets, helping to unlock substantial amounts of stranded grid power. These tools are essential for managing the energy demands and efficiency of the massive AI infrastructure. Another foundational element is NVIDIA Dynamo 1.0, an open-source software described as an “operating system” for AI inference at factory scale. Dynamo orchestrates GPU and memory resources across clusters, specifically targeting generative and agentic inference workloads, and has demonstrated significant performance boosts for inference.
Further enhancing the software capabilities are tools like NemoClaw and the broader NVIDIA AI software stack. NemoClaw is an open-source stack designed to facilitate the installation of Nemotron models and OpenShell runtime, enabling secure, always-on AI assistants. The platform also leverages the Nemotron Coalition, an initiative bringing together open model builders and developers to advance open models through shared expertise, data, and compute. Additionally, the Vera Rubin platform integrates the DOCA software framework, utilized by BlueField-4 DPUs for infrastructure services, and includes AI Enterprise software, particularly with the BlueField-4 STX storage architecture. The NVIDIA Omniverse DSX Blueprint and the Vera Rubin DSX AI Factory reference design further provide frameworks and blueprints for building, simulating, and operating large-scale, energy-efficient AI infrastructure, ensuring optimal design and deployment of these AI factories. These comprehensive software layers underscore NVIDIA’s strategy to provide a full-stack solution, from silicon to system software, for the development and deployment of advanced agentic AI. Developers building complex AI applications may find that using a vector database, such as Milvus, can complement these software layers by efficiently handling the high-dimensional vector embeddings often generated and processed within such AI factories.