The Importance of Power and Cooling in AI Implementation
In this white paper
The demand for AI is skyrocketing.
What challenges does this bring?
Rack Density
Fitting more IT gear into each rack is imperative to keep latency low and efficiency high.
More Power in smaller spaces
It's a challenge to get enough power and back-up power into each rack.
Keeping it cool
More power equals more heat. AI clusters may require advanced cooling techniques, such as liquid cooling.
The Big Picture
Remote monitoring and management of the physical infrastructure is more important than ever.
Schneider Electric supports NVIDIA AI solutions with the infrastructure you need.
Success story: Boosting AI research with unmatched efficiency
Customer challenge:
A certified multi-tenant data center customer wanted to introduce a new GPU-as-a-service offering but needed integration support.
Solution:
They engaged WWT, who helped design a solution for high-performance computing based on NVIDIA architecture. Next, they had several consultation sessions specifically for power and infrastructure, as they learned the power element was just as complex as the computing element.
For power, WWT brought in Schneider Electric. Powered by NVIDIA's HGX™ 100, which is three times faster than the previous model, the customer upgraded to a higher-voltage power distribution system to support enhanced rack PDUs. High-density computing requires robust power distribution systems to ensure reliable and efficient electricity supply.
As AI continues to demand more from data centers, IT leaders must rethink infrastructure and incorporate advanced power distribution technologies to ensure optimal performance, reliability, and energy efficiency.
Advanced computing tasks demand tightly linked GPU clusters with optimized software support. The HGX 100 is up to the challenge.
- 30x faster and 3x more energy efficient at LLM inference than its predecessor
- 3x lower TCO and requires 5x fewer server nodes
- 20x more energy efficient than CPUs for HPC and AI workloads
Schneider solutions to empower your AI data center
RACKS
- NetShelterTM SX Advanced Racks: Shock-packaged with a reinforced 4,250 lb. load rating; scaling up to 52U, 800 mm wide, and 1470 mm deep to support high-density AI servers
- 60A and 100A Rack PDUs: Designed to power high-density AI servers efficiently
- Aisle and Rack Containment: Optimizes airflow management for improved cooling and efficiency
COOLING
- Comprehensive liquid and air CDU portfolio: Solutions for both new and existing data centers
- InRow Cooling units: Precision, close-coupled air cooling for IT pods
- Air cooling accessories: Fan walls, downflow units, and rear door heat exchangers for efficient thermal management
- Rack manifolds: Simplified server integration with liquid cooling systems
POWER
- Galaxy V UPS Series: Scalable up to 1.5 MW to protect high-density AI equipment
- Remote power panels and busways: High-density power distribution solutions
- Power meters: Monitor and analyze AI-specific load profiles for optimized performance
SOFTWARE AND CONTROL
- EcoStruxureTM platform: Monitor and optimize power usage with real-time alerts and control
- EcoStruxure IT Design CFD: Cloud-based software for optimizing cooling and mitigating asset risk in data centers
- ETAP Digital Twin: A unified engineering platform for designing, operating, and analyzing electrical power systems