Thermodynamic feasibility of harvesting data center waste heat to drive an absorption chiller

Anna Haywood, Jon Sherbeck, Patrick E. Phelan, Georgios Varsamopoulos, Sandeep K. S. GuptaEnergy 2012 · DOI

TL;DR

This paper explores the thermodynamic feasibility of using waste heat from data center CPUs to power an absorption chiller, aiming to significantly improve energy efficiency. By capturing and repurposing this heat, the authors demonstrate a potential for data centers to achieve a Power Usage Effectiveness (PUE) below 1, meaning they could theoretically produce more usable energy than they consume.

What Problem Does This Paper Address?

Data centers are massive energy consumers, with a significant portion of their power dedicated to cooling the densely packed, heat-generating IT equipment. Traditional vapor-compression cooling systems are inefficient and contribute heavily to the overall energy footprint. This paper addresses the challenge of reducing this cooling energy demand by proposing a novel approach: harvesting the waste heat directly from CPUs and using it to drive an absorption chiller, a heat-driven cooling technology. This not only reduces the need for electricity-powered cooling but also recycles energy that would otherwise be lost.

Methods

The core of the methodology involves a thermodynamic analysis of a proposed system that integrates a lithium bromide-water absorption chiller with data center waste heat. Key steps include:

  • Heat Capture and Transport: Investigating methods to capture high-temperature heat from CPUs on server blades using cold plates and water as the heat transfer fluid.
  • Absorption Chiller Performance Modeling: Utilizing manufacturer specifications and a simulator tool to analyze the performance of a Yazaki WFC-SC10 absorption chiller under various temperature and heat input conditions.
  • CPU Temperature and Power Analysis: Measuring actual CPU temperatures using Dell's Open Manage System Administrator (OMSA) software and validation thermocouples to confirm they can operate within the required temperature range for the chiller. Estimating the maximum potential heat dissipation from CPUs based on their Thermal Design Power (TDP).
  • System-Level Diagram and PUE Calculation: Developing a system diagram illustrating heat and work flow and deriving a PUE equation that accounts for waste heat recovery and potential external energy sources.
  • Scenario Analysis: Evaluating PUE under different capture fractions (CF) of waste heat and coefficients of performance (COP) for both the absorption chiller and the traditional CRAC unit.

Datasets

  • CPU Temperature Data: Collected using Dell's Open Manage System Administrator (OMSA) software and validated with a National Instruments FieldPoint data acquisition device and a K-type thermocouple.
  • Yazaki WFC-SC10 Chiller Performance Data: Obtained from manufacturer specifications and a simulator tool provided by Yazaki Energy Systems, Incorporated.
  • Data Center Benchmarking Data: Referenced from industry reports and standards (e.g., The Green Grid's PUE benchmarks).

Key Insights and Ideas

  • CPU Waste Heat is a Viable Energy Source: The study demonstrates that CPUs generate sufficient high-temperature heat (up to 125°C) to potentially energize an absorption chiller, which typically requires temperatures between 70-95°C.
  • Absorption Chillers Offer a Path to PUE < 1: By effectively capturing and utilizing waste heat, the system can significantly reduce the electricity demand for cooling. When combined with external supplemental heat sources (like solar thermal), the PUE can theoretically drop below 1, indicating net energy production or export.
  • Trade-offs in Chiller Performance: There's a clear trade-off between maximizing cooling capacity and achieving peak efficiency (COP) in absorption chillers. The optimal operating point depends on the available waste heat and desired cooling output.
  • Thermal Interface Materials (TIMs) are Crucial: Minimizing the temperature difference (ΔT) between the CPU and the heat capture system is critical for efficient heat transfer. The paper highlights the importance of TIMs and proper surface preparation to reduce thermal resistance.
  • Beyond PUE: Energy Reuse Effectiveness (ERE): The authors suggest that traditional metrics like PUE may not fully capture the benefits of data centers that actively reuse energy. They point to emerging metrics like ERE that account for energy benefits extending beyond the data center's boundaries.

Key Results

  • CPU Temperatures: CPUs were shown to operate within the required 70-95°C range for chiller operation, with measured temperatures reaching up to 104°C under high load and reduced fan speeds.
  • Waste Heat Capture Potential: An estimated capture fraction (CF) of 85% of the high-quality heat (HHF) from CPUs was deemed achievable, translating to approximately 174 Wth per server blade.
  • PUE Improvement: Without external heat supplementation, the proposed system could achieve a PUE as low as 1.16, which is considered "very efficient" by industry benchmarks.
  • PUE < 1 Achievable: With external supplemental heat (equivalent to 1.05 times the IT heat), a PUE of 0.95 was calculated, demonstrating the potential for net energy production.
  • Number of Servers: The analysis indicated that even with conservative estimates, the available server blades (over 300) were sufficient to meet the heat input requirements for the absorption chiller across various operating scenarios.

Limitations and Future Work

  • Thermal Resistance (ΔT): The paper acknowledges that minimizing the temperature drop between the CPU and the heat medium is a significant challenge. While estimates were made, achieving near-perfect thermal contact and minimizing heat loss remain areas for optimization.
  • External Heat Source Integration: The study relies on the potential for external heat sources (like solar thermal) to achieve PUE < 1. The feasibility and cost-effectiveness of these supplemental sources are not fully detailed.
  • System Complexity and Cost: While thermodynamically feasible, the practical implementation of a large-scale heat capture and transport system, along with the absorption chiller, would involve significant engineering and capital investment.
  • Future Work: The authors suggest further research into developing more efficient thermal interface materials, optimizing heat exchanger designs, and exploring the integration of phase change materials for thermal storage. They also advocate for new metrics like Energy Reuse Effectiveness (ERE) to better quantify the benefits of such eco-efficient data center designs.

Summary updated

BibTeX
@inproceedings{haywood2012thermodynamic,
  author = {Anna Haywood and Jon Sherbeck and Patrick E. Phelan and Georgios Varsamopoulos and Sandeep K. S. Gupta},
  title = {{Thermodynamic feasibility of harvesting data center waste heat to drive an absorption chiller}},
  booktitle = {Energy},
  year = {2012},
  doi = {10.1016/j.enconman.2011.12.017},
}
Figure