Summary

The NVIDIA Enterprise Reference Architecture is the leading enterprise scale architecture to meet the demanding and growing needs of AI compute. The Enterprise RA represents the architecture used by NVIDIA for internal AI model training, inference for reasoning AI, and large-scale data analytics.

This specific Enterprise RA focuses on the NVIDIA B300 baseboard delivered in HGX servers. It is interconnected using NVIDIA Spectrum-X for compute and converged connectivity. It includes guidance for both dual plane and cost-effective single plane network topologies to suit a variety of deployment needs.

The Enterprise RA documents a complete system of hardware and infrastructure software necessary to implement data centers for high-performance AI functionality. It leverages decades of NVIDIA expertise in accelerated computing, networking and AI software to remove guesswork, improve speed-to-value, optimize performance and TCO, and ensure enterprise-grade reliability and security.

The Enterprise RA is designed to both support the workloads of today and grow to support tomorrow’s applications.