AIPG: Nvidia DGX Reference Architecture

The DGX H100 BasePOD environment includes 4 x H100 DGX appliances, two different 400GbE Ethernet fabrics (Cisco & Arista), 400Gb Mellanox InfiniBand fabric, and a choice of seven different storage solutions to choose from (Dell, NetApp, Pure Storage, Vast, IBM, DDN, HPE Greenlake/Vast).  The environment also includes NVIDIA's NVAIE Software Platform and Run:AI's Optimization and Orchestration solution. Customers can leverage the environment to better understand the level of effort needed to build and support their own environments along with validating the integration of different enterprise networking and storage solutions. Customers can also leverage synthetic load generation solutions to record power and performance metrics or can build their own use cases to better understand how the performance of different workloads within an integrated solution of their choosing.

This lab provides a comprehensive environment for our customers to explore, understand, and master the deployment of the Nvidia DGX H100 BasePOD for advanced AI applications. Equipped with four H100 DGX appliances and a robust selection of networking and storage solutions, this lab is designed to:

Understand DGX H100 Architecture: Participants will delve into the architecture of the DGX H100, including its integration with top-tier Ethernet fabrics (Cisco & Arista), 400Gb Mellanox InfiniBand, and a versatile array of storage solutions (Dell, NetApp, Pure Storage, Vast, IBM, DDN, HPE Greenlake/Vast).

Master Configuration and Deployment: Learn the nuances of sizing, configuring, and deploying the DGX H100 BasePOD architecture within a data center setting. This includes leveraging NVIDIA's NVAIE Software Platform and Run:AI's solutions for optimization and orchestration.

Experiment with Real-World Scenarios: Through synthetic load generation and the development of custom use cases, participants will gain hands-on experience in measuring and understanding the impact of various workloads on performance metrics within their chosen integrated solutions.

Validate Integration of Networking and Storage Solutions: The lab offers an unparalleled opportunity to test and validate the integration of diverse enterprise networking and storage solutions within an AI-driven environment, ensuring a seamless and efficient infrastructure setup.

Build and Support AI Environments: By engaging with the lab, customers will acquire the knowledge necessary to assess the effort involved in building and supporting their own AI infrastructures, fostering a deeper understanding of the complexities and requirements of cutting-edge AI deployments.

• OS/K8 Control Plane: Native K8 w/Bright Cluster Manager
• Network: Arista, Cisco, Mellanox
• Compute: Nvidia DHX H100 Appliance
• Accelerator: Nvidia H100
• Storage: Vast, Pure, Dell, IBM, DDN, Netapp