AI Reference Designs to Enable Adoption: A Collaboration Between Schneider Electric and NVIDIA

The rise of artificial intelligence (AI) is reshaping industries with applications across healthcare, manufacturing, education, entertainment, and beyond. The adoption of AI demands highly specialized hardware, specifically accelerators like NVIDIA’s GPU-based compute solutions, to handle complex tasks in AI model training. NVIDIA has evolved from a GPU supplier to an infrastructure provider, offering comprehensive solutions such as server boards, SuperPODs (AI clusters), and software stacks to accelerate AI deployment. However, integrating these high-density AI clusters into traditional data centers poses significant challenges due to the limitations of conventional power, cooling, and physical infrastructure.

Schneider Electric and NVIDIA have joined forces to address these infrastructure challenges. They have developed reference designs for both retrofitting existing data centers and building new ones specifically designed to handle the unique demands of AI. These designs focus on creating reliable, high-density setups capable of managing AI workloads while remaining energy efficient.

For existing data centers, three retrofit designs offer options depending on the cooling infrastructure and density requirements:

1. Retrofit 1: An air-cooled design accommodating 40 kW per rack, suitable for those without liquid cooling. This setup requires a larger footprint compared to liquid-cooled options.
2. Retrofit 2: A liquid-cooled design that uses existing air cooling for heat rejection, supporting up to 73 kW per rack. It balances floor space with higher density.
3. Retrofit 3: A liquid-cooled solution integrating with an existing chilled water system, also supporting 73 kW per rack but with a smaller footprint.

For new builds, Schneider and NVIDIA offer a scalable data center design optimized for AI, featuring purpose-built heat rejection systems that can handle AI clusters up to 1.8 MW with a 73 kW per rack density.

The collaboration between NVIDIA and Schneider Electric also provides valuable insights into the challenges posed by high-density AI clusters. These challenges include increased power requirements, sophisticated cooling needs (often requiring liquid cooling), and the need for fortified racks and power delivery systems. To help data center operators, the reference designs include equipment lists, physical layouts, and best practices for efficient and reliable deployment.

These reference designs offer several benefits:

• Simplified Feasibility Analysis: They help assess implementation potential, considering the limitations of traditional data centers.
• Reduced Planning Time: Pre-determined equipment lists and layouts streamline the planning process.
• Guidance on Key Challenges: The designs provide solutions for electrical issues such as short circuit currents and arc flash, and cooling complexities, particularly with liquid cooling.
• Enhanced Reliability: Rigorous validation of the designs ensures they meet high-performance requirements.
• Energy Efficiency: Sustainability is a core principle, aligning with industry goals for reducing environmental impact.

The exponential growth of AI is increasing demand for power-intensive models, with predictions of a 316% rise in AI-related power consumption by 2028. NVIDIA’s advancements, including their latest DGX models and the upcoming Blackwell platform, aim to improve energy efficiency and reduce costs, making high-performance AI more accessible.

In addition to serving as a blueprint for NVIDIA’s DGX SuperPODs, the Schneider and NVIDIA designs can be adapted for other high-density AI clusters, with considerations for specific infrastructure requirements. Organizations interested in these designs are encouraged to request detailed engineering documentation and implement best practices tailored to their needs.

Looking forward, the partnership between Schneider Electric and NVIDIA offers a robust foundation for companies adopting AI. This collaboration ensures that data centers will be able to support future AI advancements with reliable, sustainable infrastructure, empowering businesses to leverage AI’s potential with greater ease and efficiency.