Maximizing waste heat recovery from a building-integrated edge data center

Overview

Small data centres can be integrated into the energy systems of commercial and tertiary buildings to capture waste heat generated by servers, mitigate environmental impacts and enhance energy efficiency. The study introduces a methodology for maximising waste heat capture from the cooling coils by optimising workload distribution in an edge data centre of air-cooled servers. The maximisation of outlet temperatures algorithm (MOTA) was developed using a validated fast thermal evaluation approach. The pilot data centre is hosted in Empa's NEST modular research building, a mixed-use vertical energy district used in the EU projects ECO-Qube and HEATWISE.

Cooling-system parameters (air and water flow rates, coil effectiveness) were characterised experimentally. A validated multi-region CHT numerical model of the room and coil was built, with heat transfer through the coil modelled by the effectiveness-NTU method. The fast thermal evaluation (FTE) server outlet-temperature predictor had a mean bias of 1.06 °C (95% CI 0.64-1.48 °C), with Bland-Altman limits of agreement of -0.84 to 2.97 °C.

Compared with randomly distributed workloads, MOTA improves waste-heat recovery by up to 17.1%. Optimising coil water flow rate reduces the cooling load transferred to the cooling system by up to 53.2%. Higher return-air temperatures from MOTA widen the supply/return air difference, improve cooling unit efficiency and reduce the heat-pump temperature boost needed for domestic use.

Limits: validation was on a single-rack, homogeneous, air-cooled micro data centre under fixed conditions, so applicability to large heterogeneous clusters and direct-to-chip cooling is not established.

Key points

1. The pilot data centre is hosted in Empa's NEST modular research building, a mixed-use vertical energy district used in the EU projects ECO-Qube and HEATWISE.
2. The fast thermal evaluation (FTE) server outlet-temperature predictor had a mean bias of 1.06 °C (95% CI 0.64-1.48 °C), with Bland-Altman limits of agreement of -0.84 to 2.97 °C.
3. Compared with randomly distributed workloads, MOTA improves waste-heat recovery by up to 17.1%.
4. Optimising coil water flow rate reduces the cooling load transferred to the cooling system by up to 53.2%.
5. Higher return-air temperatures from MOTA widen the supply/return air difference, improve cooling unit efficiency and reduce the heat-pump temperature boost needed for domestic use.
6. Validation was on a single-rack, homogeneous, air-cooled micro data centre under fixed conditions; applicability to large heterogeneous clusters and direct-to-chip cooling is not established.

Using it in design

• In a building-integrated edge data centre, distribute workload to servers so outlet temperatures are as high as possible (within standard limits); recoverable heat rose by up to 17.1% over random distribution.
• Optimise the in-rack coil water flow rate; the cooling load transferred to the cooling system fell by up to 53.2%.
• Use the finding that higher return-air temperature reduces heat-pump temperature boost in building hot-water design, and treat results as valid for a single-rack homogeneous air-cooled layout.

Who should read it

Engineers designing waste-heat recovery and workload placement in building-integrated edge data centres.

Document contents

1. Introduction
2. Results (cooling coil characterisation, multi-region CHT model validation, FTE predictor, waste-heat recovery with MOTA, coil water flow optimisation)
3. Discussion
4. Methods (experimental setup, effectiveness-NTU method, OpenFOAM CHT model, MOTA algorithm)

Related standards

• ASHRAE TC 9.9 thermal guidelines

Read the document

Publication details

Authors

Mustafa Kuzay; Ender Demirel; Cagatay Yilmaz; Binod Prasad Koirala; Philipp Heer

Publisher

Scientific Reports (Springer Science and Business Media LLC)

Year

2025

Volume

15

Issue

1

Pages

38724

Document language

English

Source type

Scientific

Access

Open licence — downloadable

DOI

10.1038/s41598-025-22498-x

Licence

CC BY 4.0 — share with attribution

Licence evidence

doi.org

Citation