NVIDIA DGX OS

NVIDIA DGX OS is a performance-tuned Linux distribution based on Ubuntu, optimized specifically for high-density AI workloads on DGX systems.

DGX OS provides the foundational software stack for NVIDIA DGX systems (like the H100 and A100) by integrating a custom-tuned Ubuntu kernel with essential drivers and libraries. It eliminates manual configuration by pre-packaging NVIDIA Container Runtime, CUDA toolkits, and the PeerDirect storage stack for maximum GPUDirect throughput. The OS includes built-in health monitoring via the NVIDIA System Management (NVSM) tool and ensures enterprise-grade security with encrypted root filesystems and verified boot paths. This focused build delivers a stable, high-performance environment that allows data scientists to move from bare metal to model training in under an hour.

https://docs.nvidia.com/dgx/dgx-os-6-user-guide/index.html