Thermosyphon Cooler Hybrid System for Water Savings in an Energy-Efficient HPC Data Center: Results (NREL/TP-2C00-72196, September 2018)

Overview

NREL's 930 m2 (10,000 ft2) HPC data center in the ESIF building has a chiller-less design, component-level warm-water liquid cooling and waste heat reuse, and reports a trailing 12-month PUE of 1.034. Its cooling towers still consumed about 9,500 m3 (2.5 million gal) of water a year for an IT load averaging about 1 MW. In August 2016 NREL, Sandia and Johnson Controls added a BlueStream thermosyphon cooler (TSC), an advanced dry cooler using a passive refrigerant cycle, upstream of the existing towers. Tie-in needed only a one-day outage to cut the energy recovery water return pipe and fit a control valve; the unit sits on a platform meant for future towers.

Heat rejection follows a hierarchy: first, energy recovery heat exchanger to the building hot water loop; second, the TSC when temperatures allow; third, the cooling towers. Over the first full year (September 2016 to August 2017) the average IT load was 888 kW and IT energy 7,776 MWh. Heat rejection split 10.5% building reuse, 42.5% TSC and 47% towers, with the TSC dominating from November to April. Average water entering the TSC after energy recovery was 28.9 °C (84 °F). On-site WUE was 0.70 L/kWh, against 1.27 L/kWh with only heat reuse and towers, or 1.42 L/kWh with towers alone. Monthly PUE showed no negative effect, and TSC fan and extra pump energy are included in PUE.

Over two years the TSC saved 7,950 m3 (2.10 million gal), about 3.2 Olympic pools, and an estimated $5,455 in combined energy and water costs. The report stresses that NREL's decision was about sustainability rather than payback. NREL paid an average $5.77 per kgal in fiscal 2016 for water but a fixed sewer tap fee, so it saw no sewer savings. For total water including power generation, the report uses WUE source: with Colorado's energy water intensity factor of 4.542 L/kWh the first year would be 5.4 L/kWh, falling to 4.9 L/kWh after crediting a 720 kW PV system that covered 10.5% of the data center's annual energy. Cycles of concentration in the tower loop were 12.8.

The discussion explains why a TSC differs from an ordinary dry cooler. Water stays in short single-pass tubes that are easy to clean, with low pressure drop. Fan speed follows the temperature difference between entering hot water and dry bulb, and the water-to-energy cost-equivalence ratio. In cold weather refrigerant flow stops, so the loop needs no glycol and the tubes are protected from freezing. Payback drivers are high fully burdened water cost and a large gap between loop and ambient temperature. A TSC is additive: towers must still handle the full IT load in the hottest hours. Benefits include less plume and icing and more resilience. Sandia expects a single unit at 105 °F entering water to save 16,300 m3 a year in a 7-14 MW expansion.

Key points

1. In its first year the thermosyphon system saved 4,400 m3 (1.16 million gal) of water; over two years it saved 7,950 m3 (2.10 million gal), cutting data center water use by about half.
2. On-site WUE in the first year was 0.70 L/kWh, versus 1.27 L/kWh with only heat recovery and cooling towers and 1.42 L/kWh with towers alone.
3. First-year heat rejection: 10.5% building heat reuse, 42.5% thermosyphon and 47% cooling towers.
4. The data center reports a trailing 12-month PUE of 1.034, and monthly PUE showed no negative impact from adding the thermosyphon.
5. Before the TSC, the cooling towers used about 9,500 m3 (2.5 million gal) of water per year for an IT load of about 1 MW.
6. Average water entering the TSC after energy recovery was 28.9 °C (84 °F); Sandia plans 40.6 °C (105 °F) in its expansion.
7. Estimated two-year operating savings were $5,455 combined energy and water; NREL treats the project as a sustainability decision rather than a payback case.
8. Cooling tower cycles of concentration were 12.8 in the first year; NREL's water averaged $5.77 per kgal in FY2016 with a fixed sewer fee.
9. WUE source was 5.4 L/kWh with Colorado's 4.542 L/kWh water intensity factor, falling to 4.9 L/kWh after crediting 720 kW of PV.
10. A TSC is additive to the towers, which must still cover the full IT load in the hottest hours; typical use with towers can cut annual water use by up to 80%.

Using it in design

• Check whether the warm return water from liquid cooling (here 84 °F after energy recovery) leaves enough temperature difference to ambient for a dry or hybrid stage, and size it against local dry-bulb hours.
• Compare cooling options on both site WUE and WUE source whenever water savings could raise energy use, as the report recommends.
• Keep full tower capacity for peak hours when a dry stage is added, and use the dry stage to schedule tower maintenance on cool days.
• Evaluate payback using fully burdened water cost (water, sewer, treatment) and the water-to-energy cost ratio, not water price alone.
• Meter makeup, softener regeneration and blowdown separately so that WUE and cycles of concentration can be tracked; digital meters are recommended.
• Avoid glycol in the primary loop where the dry device can protect itself against freezing, since glycol reduces heat transfer.

Who should read it

Designers and operators of liquid-cooled or warm-water data centers weighing dry, hybrid and evaporative heat rejection against site water constraints.

Document contents

1. Executive Summary
2. 1 Introduction
3. 1.1 Thermosyphon Installation: A Brief Recap
4. 1.2 ESIF HPC Data Center Cooling System: A Brief Recap
5. 1.3 Data Center Metrics: PUE, ERE, and WUE
6. 2 Results
7. 2.1 Heat Rejection Options: Sample Data
8. 2.2 First Year of Thermosyphon Operation: Data Center Metrics
9. 2.3 Two Years of Thermosyphon Operation: Cumulative Savings
10. 2.4 Performance Notes
11. 3 Further Discussion
12. 3.1 Thermosyphon: Additional Information
13. 3.2 ESIF Water Information
14. 3.3 Sandia Data Center Expansion: TSC Plans
15. 4 Conclusion

Publication details

Authors

David Sickinger; Otto Van Geet; Suzanne Belmont; Thomas Carter; David Martinez

Publisher

NREL

Year

2018

Document language

English

Source type

Government / public

Access

Citation only

Licence

Licence conditions not established — confirm the conditions with the publisher

Licence evidence

www.nrel.gov

Citation