Liquid Temperature Control for AI Servers (DLC)
Authored by: Dr. Dirk Frese, VP of Sales, Marketing & Service
Semiconductors in chips need a temperature between 0 °C and 70 °C for commercial or -40 °C to 85 °C for industrial applications. For the use of servers for AI or crypto mining applications, a temperature between 17 °C and 30 °C is generally recommended. The heat generated by this application is enormous, and simple HVAC systems to climatize server rooms are not enough anymore.
AI server banks have an even higher heat load than crypto mining farms, as the allowable downtime is usually less than 1% compared to 5% for crypto mining. Therefore, we are reading about small nuclear power plants as the future solution to power AI servers at large scale.
To effectively cool AI servers, direct liquid cooling (DLC) has become more important than cooling the room they’re located in. This is much more efficient to dissipate heat from the chips quickly and in a stable fashion. However, it adds to the complexity of wiring and tubing.
Using circulators and liquid temperature control for AI servers is critical for maintaining optimal performance and hardware longevity. Here are some key points about their application:
Precision Temperature Control
Circulators maintain precise and uniform temperatures of liquid coolants, which is essential for the stability and reliability of AI servers. This precision helps mitigate thermal effects and background noise, enhancing the accuracy and sensitivity of computational processes and data analysis.
Direct Liquid Cooling (DLC) Systems
Direct Liquid Cooling (DLC) systems involve direct contact between heat-generating components (e.g., CPUs, GPUs) and cold plates connected to tubes with coolant inside. This method effectively removes heat from the server more efficiently than traditional air-cooled systems, ensuring overall system performance and stability. DLC systems can be integrated with both liquid-to-air and liquid-to-liquid coolant distribution units (CDUs), making them adaptable to existing data center infrastructures.
Photo Credit: Shutterstock
Benefits of Liquid Cooling
1. Higher Efficiency: Liquid cooling transfers heat much faster than air cooling, which is crucial for handling the high-density computing requirements of AI workloads.
2. Space Optimization: By optimizing space and decreasing component failure rates, liquid cooling systems help achieve sustainable and performance objectives in cloud computing and scientific computing environments.
3. Scalability: Liquid cooling solutions can be scaled to meet the increasing demands for greater heat dissipation, which is essential for AI servers that handle complex and heavy computational tasks.
Applications in Data Centers
Liquid cooling solutions are increasingly adopted in data centers to support high-performance computing (HPC) clusters. These solutions include cold plate liquid cooling systems, which maximize the utilization of liquid-cooled plate heat exchange to cool the main heat sources inside the server, thus reducing the use of traditional cooling systems and improving power efficiency.
JULABO and Server Cooling
Julabo’s precise liquid temperature control solutions cover a wide range of cooling capacities using electrothermal or Peltier cooling, as in the TE400 product with 400W of cooling power or air-cooled as well as water-cooled products with cooling capacities of up to 23.5kW (FL20006 and FLW20006). This means that the cooling and stability needs are met from R&D test set-up to larger installation.
Conclusion
Using circulators with liquid temperature control for AI servers is a sophisticated and efficient method to manage the thermal demands of high-density computing environments. By ensuring precise and stable temperature control, these systems enhance the performance, reliability, and longevity of AI servers, making them indispensable in modern data centers.
Citations:
Ramakrishnan, Bharath, et al. “Experimental characterization of two-phase cold plates intended for high-density data center servers using a dielectric fluid.” Journal of Electronic Packaging 143.2 (2021): 020904.
Ramakrishnan, Bharath. Experimental Assessment of Liquid Cooled Cold Plates Intended for Improving the Operational Efficiency of a High
Density Data Center Environment. Diss. State University of New York at Binghamton, 2019.
Hoang, Cong Hiep, et al. “Two-Phase Impingement Cooling Using a Trapezoidal Groove Microchannel Heat Sink and Dielectric Coolant HFE 7000.” 2021 20th IEEE Intersociety Conference on Thermal and Thermomechanical Phenomena in Electronic Systems (iTherm). IEEE, 2021.
