NVIDIA AI Enterprise Expands with H200 NVL GPU Support

NVIDIA has announced a significant update to its NVIDIA AI Enterprise platform, now incorporating support for the NVIDIA H200 NVL GPU. This advancement is part of the latest release of the company’s infrastructure software, aimed at enhancing enterprise-level AI applications. The H200 NVL, a new addition to NVIDIA’s data center GPU lineup, promises to deliver cutting-edge capabilities for agentic and generative AI, according to NVIDIA.

NVIDIA AI Enterprise Platform

The NVIDIA AI Enterprise platform is designed to facilitate the development and deployment of production-grade AI solutions. It consists of a comprehensive suite of software components that can be deployed on various hardware setups, including servers, edge systems, and workstations. The platform is divided into two main categories: the AI and Data Science software catalog and the infrastructure software collection.

The AI and Data Science software catalog features NVIDIA NIM microservices and several frameworks for building AI workflows. These components are containerized for seamless cloud-native deployment, ensuring compatibility with various cloud service providers.

Infrastructure Software Collection

The infrastructure software collection provides essential components for supporting AI and data science workloads on accelerated systems. This includes drivers for GPU, networking, and virtualization, as well as Kubernetes operators. Additionally, the Base Command Manager Essentials is available for efficient cluster management.

With the latest update, the infrastructure software collection now supports the H200 NVL GPU, a move expected to significantly boost AI application performance and energy efficiency.

H200 NVL GPU Enhancements

Revealed at the Supercomputing 2024 conference, the H200 NVL GPU is designed for data centers requiring lower-power, air-cooled enterprise rack designs. It offers flexible configurations to accelerate a wide range of AI workloads. The GPU boasts a 1.5x memory increase and a 1.2x bandwidth increase over its predecessor, the NVIDIA H100 NVL, delivering up to 1.7x faster inference performance.

Support for the H200 NVL in NVIDIA AI Enterprise is being rolled out in phases. Version 6.0 of the infrastructure collection, available now, supports bare-metal applications and virtualization with GPU pass-through. Version 6.1, expected later, will add support for virtualization with vGPU.

Reference Architecture and Availability

NVIDIA has also introduced a reference architecture to streamline the deployment and configuration of AI systems. This architecture provides a flexible infrastructure stack for original equipment manufacturers (OEMs) and partners, ensuring consistent software components and adaptable hardware configurations.

For enterprises purchasing servers with the H200 NVL, NVIDIA AI Enterprise is immediately accessible, complete with a five-year subscription. Additionally, NVIDIA offers several ways to get started, including free access to NIM microservices for testing and a 90-day free evaluation license. The NVIDIA AI Enterprise Infrastructure Collection 6.0 is available for download from the NVIDIA NGC Catalog.

Image source: Shutterstock