Vertiv & NVIDIA launch architecture for AI data centres

Vertiv has announced the release of a 7MW reference architecture designed in collaboration with NVIDIA for the NVIDIA GB200 NVL72 platform.

This new architecture aims to transform traditional data centre methods into facilities capable of supporting artificial intelligence (AI) applications enterprise-wide. By co-developing with NVIDIA, Vertiv has created a blueprint intended to expedite the deployment of the liquid-cooled, rack-scale NVIDIA GB200 NVL72 platform, with each rack supporting up to 132kW. The architecture focuses on optimising deployment speed, performance, resiliency, cost, energy efficiency, and scalability.

"We are proud to deepen our collaboration with NVIDIA to enable AI-driven data centres of today and tomorrow," said Giordano (Gio) Albertazzi, CEO of Vertiv. "As a leader in critical power and cooling infrastructure, Vertiv is uniquely positioned to support the NVIDIA GB200 NVL72 platform. Our portfolio of high-performance power and cooling solutions, combined with our global scale, will enable customers to deploy AI data centres faster, more efficiently and with greater flexibility to address densification, dynamic workloads, retrofits, and enable future-ready designs."

Jensen Huang, founder and CEO of NVIDIA, stated: "New data centres are built for accelerated computing and generative AI with architectures that are significantly more complex than those for general-purpose computing. With Vertiv's world-class cooling and power technologies, NVIDIA can realise our vision to reinvent computing and build a new industry of AI factories that produce digital intelligence to benefit every company and industry."

The architecture is intended to simplify and accelerate the deployment of AI workloads in both existing and new data centres while reducing risk and enhancing standardisation. The designs incorporate critical power infrastructure that aligns AI clusters with data centre capacity blocks. The infrastructure effectively manages high-density heat removal through a hybrid liquid- and air-cooling system. Optional systems inspired by the Open Compute Project (OCP), such as DC power shelves, are also included in the guidance provided.

Vertiv's 360AI portfolio offers retrofit and greenfield reference designs to guide customers with integrated power and cooling solutions for AI and high-performance computing. The key features of the design for the NVIDIA GB200 NVL72 include rapid deployment and retrofitting using preconfigured modules, space-saving power management technologies, and efficient cooling solutions. Additionally, the architecture supports dynamic workloads and comes with extensive installation and operational services.

Key benefits of this co-developed reference architecture include the rapid deployment and retrofit potential, up to 50% faster than traditional builds, space-saving power management that offers enhanced reliability and efficiency, and energy-efficient cooling technologies which can reduce cooling costs by up to 20% compared to previous solutions. It also includes dynamic workload management and services from thousands of field engineers worldwide.

The collaboration between Vertiv and NVIDIA aims to provide a roadmap for technical co-development and enable the large-scale deployment of accelerated computing. Experts are made available to discuss the design in more detail at industry events.