Summary:
1. Parker Hannifin experts discuss the use of direct-to-chip liquid cooling, careful routing, and advanced monitoring to manage high heat loads in AI data centers.
2. AI data centers require more power and generate more heat, leading to the need for compact cooling solutions to preserve rack density.
3. Two-phase liquid cooling systems are becoming popular for high-power electronics due to their efficiency in managing heat.
Rewritten Article:
In the realm of AI data centers, the challenge lies in managing the escalating power requirements and heat generation of new AI workloads. To address this, Parker Hannifin’s Elvis Leka and Josh Coe delve into the realm of direct-to-chip liquid cooling, meticulous routing, low-restriction couplings, and sophisticated monitoring and control systems. These strategies work together to effectively handle high heat loads while maintaining rack density and system efficiency.
The demand for more power in AI data centers has resulted in higher heat generation, necessitating additional cooling solutions. However, the irony lies in the fact that AI server racks, designed to handle increased power requirements, are denser than standard compute racks, leaving limited space for cooling systems. The compact nature of cooling systems becomes crucial to avoid consuming valuable floor space and to allow for maximum rack density and future expansion.
Moreover, the concentrated heat zones, or hotspots, created by high-density AI servers require targeted cooling solutions placed directly next to them. Industry-specific compact cooling systems play a vital role in delivering thermal management precisely where it’s needed, ensuring efficiency in the ecosystem and preventing localized temperature issues that could lead to premature equipment failure.
AI workloads demand higher power density per rack, with estimates suggesting they can be significantly greater than traditional data centers. The thermal design power (TDP) of AI chips continues to rise, leading to space constraints when integrating advanced cooling systems. The use of highly dense clusters of powerful chips, such as GPUs, in AI data centers puts traditional cooling systems under immense pressure, necessitating more powerful and energy-intensive cooling solutions.
To combat the challenges posed by high-density AI workloads, the industry is transitioning towards liquid cooling options, including immersion cooling and two-phase cooling. Liquid cooling offers benefits such as enhanced heat transfer efficiency, reduced costs, and a smaller environmental footprint. By significantly lowering energy consumption, liquid cooling not only cuts down operational expenses but also contributes to a more sustainable data center environment.
Direct-to-chip liquid cooling systems, categorized into single-phase and two-phase variants, involve mounting a cold plate heat exchanger directly onto energy-dense components like CPUs and GPUs. While single-phase liquid cooling systems suffice for current data centers, the increasing power demands of future AI-rich data centers are leading to a rise in popularity of two-phase options.
Two-phase cooling, where heat absorption occurs through latent heat during the refrigerant’s phase change, is ideal for high-power electronics that surpass the capabilities of traditional air and water cooling systems. These systems efficiently transfer high heat and disperse latent heat generated by electronic components with heightened power densities, offering a streamlined design with excellent thermal performance/cost ratio.
However, direct-to-chip pumped two-phase systems come with challenges, including higher upfront investment requirements and the need for specialized maintenance training. Designing a system that effectively controls flow and pressure across cooling loops poses a technical hurdle that requires careful consideration.
As the efficiency-driven nature of newer two-phase liquid cooling systems necessitates precise temperature, pressure, and flow control, advanced monitoring and control systems are indispensable. These systems rely on sensors and AI/ML-driven analytics to monitor conditions in real-time, make dynamic adjustments to offset workload fluctuations, and enable predictive maintenance.
Efficiency is paramount in cooling system design, with Power Usage Effectiveness (PUE) playing a crucial role in assessing energy usage at different levels within a data center. Lowering PUE values across server blade, IT rack, and facility levels enhances overall energy efficiency and sustainability, ensuring that power is utilized by IT equipment rather than being wasted on cooling and power delivery.
To maximize efficiency, careful consideration must be given to sizing requirements for all components in the cooling system. Properly sizing each component to match the thermal load, minimizing energy consumption, and optimizing heat transfer is essential for enhancing system performance while avoiding energy wastage and potential system failures.
Pressure drop, the decrease in fluid pressure as it flows through a system, is a critical factor in liquid cooling system design. Excessive pressure drop can lead to reduced system efficiency, increased energy costs, and accelerated equipment wear. Strategies for minimizing pressure drop involve optimizing system design, selecting appropriate components, and implementing maintenance practices to ensure efficient fluid flow and system operation.
In conclusion, the evolving landscape of AI data centers calls for innovative cooling strategies to address the escalating power demands and heat generation. By embracing compact, efficient cooling solutions and optimizing flow rates and routings, the industry is paving the way for improved component longevity, reduced operational costs, and enhanced system performance in the face of high-density AI workloads.
