Research on Design and Optimization of Economic Operation for Indirect Liquid Cooling System in Data Center Servers

Overview

Cooling system energy is cited as accounting for over 30% of data-centre energy consumption. Liquid cooling shows superior cooling performance but often leads to high system power consumption because of design and flow-matching factors, so energy-saving optimisation of liquid cooling systems is important. The modelled cabinet has five liquid-cooled servers connected in parallel, each represented by a cold-plate-to-fluid thermal resistance in an equivalent network. The heat-flow method and piping-network fluid dynamics resistance-balance relationships establish the system flow and heat-transfer constraints.

With a centralised pump and constant heat load, raising cooling water temperature from 20 °C to 24 °C increases total power consumption by 1.55 times. At constant total pump power, a 2 °C rise in cooling water inlet temperature reduces the removable thermal load by 4.9%. Placing a 1.2 kW server at the bottom of the cabinet instead of the top lowers total power consumption by 34.4%, and higher-load servers should sit near the supply. As total thermal load rises from 4.0 kW to 6.0 kW, a distributed pump configuration saves on average 2.5 W (maximum 7.09 W) versus a centralised pump.

The optimisation finds the optimal pump frequency first to minimise pump power at fixed heat load and then to maximise heat load at fixed pump power. Facility-side cooling water temperature and server position in the cabinet both strongly affect pumping energy, while distributed pumps give only watt-level savings at this scale.

Limits: a single five-server cabinet model with piping-network assumptions; results depend on this cabinet layout and pipe design. Savings figures are at cabinet scale, and the distributed pump benefit is small.

Key points

1. Cooling system energy is cited as accounting for over 30% of data-centre energy consumption.
2. The modelled cabinet has five liquid-cooled servers connected in parallel, each represented by a cold-plate-to-fluid thermal resistance in an equivalent network.
3. With a centralised pump and constant heat load, raising cooling water temperature from 20 °C to 24 °C increases total power consumption by 1.55 times.
4. At constant total pump power, a 2 °C rise in cooling water inlet temperature reduces the removable thermal load by 4.9%.
5. Placing a 1.2 kW server at the bottom of the cabinet instead of the top lowers total power consumption by 34.4%; higher-load servers should sit near the supply.
6. As total thermal load rises from 4.0 kW to 6.0 kW, a distributed pump configuration saves on average 2.5 W (maximum 7.09 W) versus a centralised pump.

Using it in design

• In in-cabinet liquid cooling, place high-load servers near the supply side; locating a 1.2 kW server at the bottom cut total power by 34.4%.
• Weigh facility-side cooling water temperature together with pump energy and chip margin; going from 20 to 24 °C raised total power by 1.55 times.
• Treat the distributed-pump decision as a small cabinet-scale gain (average 2.5 W) to be weighed against cost and complexity, and optimise pump frequency for minimum power at fixed load and maximum heat load at fixed power.

Who should read it

Engineers designing cabinet-level indirect liquid cooling loops, pump sizing and operating optimisation.

Document contents

1. 1. Introduction (1.1 Background of Data Center Thermal Management: current status of air cooling, liquid cooling, energy-saving optimization; 1.3 The Main Work in This Paper)
2. 2. Analysis of Piping Network Resistance Characteristics in Server (2.1 Introduction to Cabinet-Level Liquid Cooling Systems and Piping Network Design, 2.2 Piping Network System Design and Resistance Calculation)
3. 3. Heat Flow Model for Cabinet Indirect Liquid Cooling Systems (3.1 Analysis of Heat Transfer and Flow Characteristics, 3.2 Boundary Conditions)
4. 4. Comprehensive Economic Operation Optimization Analysis of Server (4.1 Economic Operation Optimization of Centralized Pump Systems: under given total load and given total power; 4.2 Economic Operation Optimization of Distributed Pump Systems)
5. 5. Conclusions

Read the document

Publication details

Authors

Yuxuan Xin; Daoguang Yu; Xiaohan Ren

Publisher

Energies (MDPI AG)

Year

2026

Volume

19

Issue

13

Pages

3068

Document language

English

Source type

Scientific

Access

Open licence — downloadable

DOI

10.3390/en19133068

Licence

CC BY 4.0 — share with attribution

Licence evidence

www.mdpi.com

Citation