DCFTS 2024 Session 4 - The Future of AI Data Center Design: Power, Cooling and Chip Considerations

Dec. 9, 2024
Here's an audio recording of Session 4 from the inaugural Data Center Frontier Trends Summit, held Sept 4-6, 2024, in Reston, Virginia.

Session Description from 2024 DCF Trends Summit Program:

The Future of AI Data Center Design: Power, Cooling and Chip Considerations

The average rack density in most data centers today is 7-10 kW. Modern AI workloads are driving deployments of GPUs and other AI infrastructure that can require 50 kW today and, in fact, well over 100 kW in a single rack. The rapid acceleration in power densities and performance needs are driving change in data center facility design, and the need for more sustainable infrastructure.

This session will cover the theoretical and practical design limitations to support AI, and solutions to engineer around them. We will discuss how a data center is designed and built, how the use cases that the infrastructure deployed in them supports help shape that design, and what the AI data center looks like from a construction, cooling, and connectivity perspective.

The session will also look into the variety and nuances of GPU chip considerations in data centers with relation to power density and liquid cooling technology options.

Sponsored Recommendations

Optimizing AI Infrastructure: The Critical Role of Liquid Cooling

In this executive brief, we discuss the growing need for liquid cooling in data centers due to the increasing power demands of AI and high-performance computing. Discover how ...

AI-Driven Data Centers: Revolutionizing Decarbonization Strategies

AI hype has put data centers in the spotlight, sparking concerns over energy use—but they’re also key to a greener future. With renewable power and cutting-edge cooling, data ...

Bending the Energy Curve: Decoupling Digitalization Trends from Data Center Energy Growth

After a decade of stability, data center energy consumption is now set to surge—but can we change the trajectory? Discover how small efficiency gains could cut energy growth by...

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

Traditional data center power, cooling, and racks aren’t sufficient for GPU-based servers arranged in high-density AI clusters...