Accelerating AI Deployments Through Collaborative Problem Solving
The adoption of artificial intelligence is creating a seismic shift in the data center landscape, impacting design and operations for enterprises, colocation providers, and hyperscale giants alike. With over 30 years of experience serving multinational customers, we’re at the forefront of the AI transformation, collaborating with our customers to develop practical solutions that are not only forward-thinking, but immediately solve the evolving challenges of deploying resilient, agile AI data centers.
Reference Designs: Supporting the Nvidia DGX H100 SuperPOD
See Legrand’s own Reference Design, detailing the specific Legrand components across power, cabinets, thermal management, and connectivity.
Review Nvidia’s Reference Design, which includes specific inclusion of Raritan and Server Technology PDUs and Legrand cabinets.
Meet Rising Power Needs with Safety and Efficiency
With high-performance GPUs and CPUs demanding over 100kW per rack, lower power capacity rack PDUs and traditional pipe and wire or conduit are no longer sufficient. Explore how intelligent, high-amperage PDUs and overhead track busway meet rising power supply needs safely and efficiently.
Leverage advanced rack PDU solutions to meet the most demanding power requirements. With high density PDUs ranging from 208 to 480V and 30A to 125A, we can meet rack capacity of 10kW to 90KW. Customize form factor, outlet configuration, outlet type and more.
Utilize track busway to adapt to the ever-changing power needs while building in redundancy and efficiency. Starline’s T5 track busway provides up to 1250 amps per run, and is rated up to 600V. AI deployments can utilize multiple overhead runs to supply up to 6000A.
Mitigate risk and increase safety with Starline Remote Plug-in Actuators. These devices allow you to add or remove connections from a live powered busway from outside the arc flash boundary.
Efficient Cooling Options to Support Processing Demands
Rising power density and increased computational power significantly raise cooling requirements, with traditional air cooling unable to keep pace. Rear door heat exchangers and modular containment options enable additional thermal management without overhauling your infrastructure for liquid cooling.
USystems’ reardoor heat exchangers utilize air-assisted liquid cooling to support deployments of over 150kW per rack. This efficient cooling solution has a potential cooling PUE of 1.035, and and is compatible with open and closed water loop systems. As an easy addition to Legrand cabinets, they expand cooling capability while eliminating the need for larger reconfigurations.
Maximize efficiency and reliability with Contain-IT FLEX. This modular system supports both hot aisle and cold aisle in the same deployment. It ensures consistent airflow and temperature, increasing efficiency and enabling cooling infrastructure to handle higher densities.
Managing airflow at the cabinet level is an essential step to energy efficiency. Select cooling-agnostic Legrand customizable cabinets, which provide compatibility with heterogeneous cooling strategies like Liquid-to-air and Liquid-to-liquid cooling options in your AI data center.
Reimagining the Rack: AI Requirements Change Physical Specs, Deployment Expectations
The ability of AI to process massive amounts of data hinges on high-density clusters. Standard racks were not designed to support the larger size or heavier weight loads of these clusters. At the same time, there is growing pressure to increase speed to deployment with pre-integrated, ready-to-install racks.
The T6 and T-Series are engineered and tested for high-density deployments. Built on a fully welded, tubular steel frame, they are rated for 5,000 lb static and 4,000 lb dynamic loads. Design the cabinets to suit your specific dimensions in heights up to 52U.
Our T6 and T-Series cabinets are also available as a Rack and Stack solution with equipment pre-loaded, helping shorten time to deployment. Tested to hold up to 3,500 lbs (ISTB 3B), and utilizing circular packaging materials, they’re a sustainable choice for high-density applications.
Overcoming Challenges in Cable Management and Connectivity
Cabling requirements for Al servers, especially GPU clusters, can impede proper airflow and reduce cooling efficiency. Additionally, complex cable routing can cause delays in installations. Efficient and well-planned cabling are crucial to ensure maximum performance and low latency while utilizing minimal space.
Our Approved Networks optical networking solutions offer the highest levels of performance with high-speed data rates up to 800G. Transceiver channelization (400G/800G transceivers) with Very Small Form Factor (VSFF) connectivity further supports a massive scaling down in cabling.
Featuring the industry’s lowest optical loss by a significant margin, acclAIM works with both front-end compute networks or AI back-end and pairs perfectly with transceiver channelization. It features multiple uses for virtually infinite scalability and adaptability.
With a wide array of managers, panels, shelves and enclosures, you can combine accessories for a nearly custom solution to organize cables and ensure proper airflow.
Critical Monitoring and Management for AI Deployments
Higher power densities necessitate more granular power monitoring, both for usage, and for quality. Simply put, the risks of not uncovering power inefficacies your AI deployment are too great to ignore. Reliable, real-time usage visibility empowers data center operators to meet their efficiency, sustainability, reporting and management goals, while power quality monitoring ensures smooth operation and prevents data loss.
Overseeing AI workflows can be further challenged by deployments in multiple geographic locations utilizing varios operating system languages. Out-of-band (OOB) management should be considered for simplifying systems-level visibility and control.
As power density increases, Legrand’s SmartSensors help maintain optimal conditions through precise monitoring of temperature, humidity, airflow, vibration, water/fluid leaks, and other key environmental metrics.
Rack power monitoring with intelligent PDUs including the Raritan PX4 and Server Technology PRO4X enables granular visibility at the outlet or device level, with advanced power quality metrics that help AI data center maintain uptime and perform issue analysis.
The Starline M70 adds revenue-grade power monitoring at multiple levels along the busway, including at the power feed and branch-circuit. The real-time data helps maintain peak efficiency and balance power loads, crucial for HPC.
Protect your multimillion-dollar AI investment with out-of-band management as recommended by Nvidia's Reference design. ZPE Systems' serial controllers are the best practice for resilient, secure, and automated infrastructure for ZTP, daily operations and dooms day recovery.