NVIDIA has announced the NVIDIA OpenShell Reference Architecture, a full-stack infrastructure for building, containing, observing and enforcing safe agents across enterprise infrastructure and edge systems. For agentic workloads, NVIDIA Vera Rubin NVL72 systems deliver 30x higher throughput per megawatt.
The architecture combines NVIDIA Vera Rubin infrastructure for model inference, NVIDIA Vera CPUs for agent harnesses and tool use, open source NVIDIA OpenShell secure runtime software, and NVIDIA BlueField-4 DPUs running NVIDIA Sentry for monitoring and in-silicon policy enforcement.
NVIDIA says Vera CPUs deliver 2x the performance of x86 processors with greater energy efficiency. BlueField-4 DPUs continuously monitor agent behavior with NVIDIA Sentry, built on NVIDIA DOCA software. If an agent attempts to move outside its software boundary, Sentry quarantines and stops it in milliseconds; NVIDIA says the software boosts agent capacity by up to 20%.
The architecture extends to edge systems including NVIDIA RTX Spark, NVIDIA DGX Spark and NVIDIA DGX Station. It runs on enterprise operating systems and container orchestration layers including Linux and Kubernetes infrastructure powered by Canonical, SUSE and Red Hat, as well as Microsoft Windows using Microsoft Execution Containers.
“AI capability and safety must advance together,” said Jensen Huang, founder and CEO of NVIDIA.
The NVIDIA OpenShell Reference Architecture is now available. Details are at https://nvidianews.nvidia.com/news/openshell-reference-architecture.
