Mouser - White Papers

Mouser Electronics White Paper Artificial intelligence (AI) has changed the physical reality of the data center. Training clusters and high-performance inference systems pack a significant number of graphics processing unit (GPU) cards and accelerators into compact rack spaces. Along with this, increasing chipset performance has raised electrical demand, and nearly all of that energy ultimately becomes heat that must be managed. This management is critical, as stable operating conditions determine whether data centers deliver sustained performance or encounter throttling, downtime, or premature failure. Managing this thermal load, however, is becoming more expensive, as cooling infrastructure and power distribution networks account for a substantial share of both capital expense and operating costs. A large portion of facility energy is spent on cooling rather than IT systems. In fact, cooling infrastructure can consume 30 to 40 percent of the total energy budget, with less efficient facilities exceeding that range. 1 Rack power density is moving from tens of kilowatts toward several hundred kilowatts, with some deployments approaching 1MW. Rack performance is now constrained as much by environmental control as by computing power. With heat and power management becoming part of data center system design, rather than just a facilities concern, this white paper examines engineered thermal, sensing, and connector solutions for high-density compute systems. Designing for Efficiency Within Recognized Standards A data center's efficiency directly affects operations. Equipment must remain within temperature and humidity envelopes specified by manufacturers, while total facility energy consumption remains controlled. Industry standards define the measurable efficiency targets. The American Society of Heating, Refrigerating, and Air- Conditioning Engineers (ASHRAE) Technical Committee 9.9 (TC 9.9) defines the recommended environmental ranges for data processing equipment and includes guidance on sensor deployment. 2 There is also the ASHRAE Standard 90.4, which addresses the role of energy efficiency in data centers and supports the use of continuous monitoring tied to control systems. 3 In addition to international standards, regional regulations add further requirements. For example, California's Title 24 demands the use of monitoring systems for heating, ventilation, and air conditioning (HVAC) and lighting in certain jurisdictions. Broader frameworks, such as EN 50600, ANSI/TIA-942, and ISO 14001, offer guidance on infrastructure, operations, and environmental management to build reliable and sustainable data centers. 4 Accurate measurements are a crucial part of each of these standards. High-density AI systems create heat that moves with the workload, so fixed assumptions about room temperature and airflow are unreliable. Meeting efficiency targets requires more than just knowing the conditions—it requires systems that can respond to them. HVAC Feedback Loops and Precision Monitoring Cooling performance depends on feedback. Sensors report temperature and humidity, while control systems adjust the flow of air, chilled water, or cooling liquid in response to this feedback. Without continuous data, cooling systems either overcompensate and waste energy or underperform, exposing equipment to thermal stress. Room-level measurements alone cannot capture the full picture. Within a single rack, vertical gradients can develop. The air entering at the front may be within the expected temperature range, but exhaust air at the rear could exceed recommended limits. Additionally, restricted airflow areas can trap heat around power supplies or accelerator modules, while raised floor cavities may conceal localized hot zones. For monitoring to be effective, sensors must be distributed throughout the installation. Engineers should position temperature sensors at different heights within each rack, near high-power components, and in regions where airflow paths narrow. Dispersing feedback capture in this way ensures monitoring extends beyond room-level measurements to include the equipment itself. That kind of distributed monitoring depends on sensors that can be placed close to the equipment without getting in the way. Amphenol Advanced Sensors JI and JIC waterproof IP68 negative temperature coefficient (NTC) thermistors support this model. They have a compact format that allows for installation in confined spaces, such as server enclosures, switchgear compartments, and cooling assemblies. Meanwhile, the IP68 sealing protects against condensation or moisture. When connected to building management or rack-level controllers, these thermistors provide stable measurements that strengthen closed-loop HVAC control. The Transition to Liquid Cooling Conventional computing relies on airflow, heatsinks, and fans for cooling. But higher rack densities have forced engineers to revisit these approaches. Air cooling is effective at modest power levels, but as power densities increase, its limitations

• Cover