Data Center Heat Rejection and the 43 °C Dry Cooler Ceiling

A dry cooler leaves its liquid 5.5 °C (10 °F) above the outdoor dry-bulb temperature, according to the approach table in “Hot” for Warm Water Cooling, a paper by Lawrence Berkeley National Laboratory, IBM and Sandia National Laboratories presented in 2011. Add that to the hottest design dry-bulb among the paper’s U.S. sites, 37.5 °C, and the warmest liquid a dry cooler can be counted on to supply is 43 °C. Every exchanger between the cold plate and the outdoor coil takes more kelvin off what the chip can be given. Run the same sum for a current coolant temperature and the outdoor air has to be far cooler.

Three loops, one approach penalty per hop

Schneider Electric’s White Paper 210 redraws an ASHRAE diagram that splits a liquid-cooled hall into three water systems. The technology cooling system (TCS) carries coolant through the cold plates. A coolant distribution unit that separates the two loops (CDU) passes that heat to the facility water system (FWS), and the FWS runs to the outdoor heat rejection equipment. Where a chiller and a cooling tower do that job, the diagram adds a condenser water loop between them. The paper lists the outdoor options as air-cooled or water-cooled chillers, a dry cooler, and a dry cooler with a water spray, which it calls adiabatic cooling.

The cold plates do not catch everything. The 2011 paper puts the liquid’s share at 90 percent or more of the heat. The rest reaches room air, and Schneider describes halls where one chiller plant serves both the liquid-cooled and the air-cooled load.

Each hop needs a temperature difference to move heat, called an approach. For a cooling tower, the Energy Design Resources chilled water guide defines it as the leaving water minus the ambient wet-bulb temperature.

Its nominal tower is rated at 95 °F entering water and 78 °F entering wet-bulb. The water leaves at 85 °F, so the approach is 7 °F.

The 2011 paper assigns 3.8 °C (7 °F) to an open cooling tower, which agrees with the guide, and 1.67 °C (3 °F) to a plate-and-frame exchanger between tower water and building liquid.

A dry fin cooler gets 5.5 °C (10 °F) and a CDU gets 2.77 °C (5 °F).

Schneider’s paper gives current CDU approaches of 2 °C to 4 °C, with some vendors as high as 6 °C, so the 2011 CDU value still sits inside the range. Coil values depend on coil size: the 2011 paper says a bigger dry cooler gives a smaller approach at higher cost and more space.

Source: Coles et al., “Hot” for Warm Water Cooling (2011), Table 2; Schneider Electric, White Paper 210 (2025), Figure 1; Energy Design Resources, Chilled Water Plant Design Guide (2009).

A 43 °C ceiling for dry coolers

The 2011 paper adds approaches in series from the outdoor design condition. For a tower path the sum is 26.5 °C wet-bulb + 3.9 K tower approach + 1.7 K plate-and-frame approach = 32.1 °C, which the paper reports as 32 °C.

For a dry cooler the sum is 37.5 °C dry-bulb + 5.5 K = 43.0 °C.

Both results are upper bounds for the design hour, which the paper sets at 99.6 percent of hours, all but 35 per year.

Table 1 runs the same two sums for five of the paper’s sixteen locations. The gap between the paths follows humidity rather than heat. Los Alamos opens a 16.5 K gap and Newport News only 8.1 K, though their dry-bulb values differ by 0.4 K.

Table 1. Liquid supply at the 99.6 percent design hour, by outdoor exit (°C; gap in K)

* Supply = design dry-bulb + 5.5 K (dry cooler); design wet-bulb + 3.9 K + 1.7 K (tower and plate-and-frame exchanger)

* Rows sorted by dry cooler supply; Gap = dry cooler supply minus tower supply; locations as named in the paper, nearby ASHRAE stations

Source: Coles, Ellsworth and Martinez, “Hot” for Warm Water Cooling (2011), Tables 2 to 4 and Appendix A; supply and gap columns calculated here

Source: Coles et al., “Hot” for Warm Water Cooling (2011), Tables 2 to 4 and Appendix A; supply columns calculated here.

The 2021 ASHRAE TC 9.9 white paper names the water classes by their upper supply limit: W17, W27, W32, W40, W45 and W+. Against those classes the dry cooler’s 43 °C misses W40 and clears W45 by 2 K.

Counted site by site with the same approach, nine of the 16 locations let a dry cooler hold 40 °C or lower, and none lets it hold 32 °C.

The tower path serves W32 at all but two of the 16 locations. The misses are Newport News, by 0.1 K, and Houston, whose 26.7 °C wet-bulb exceeds the 26.5 °C ceiling the paper applies to every site. Houston was added as a location of interest, so the paper’s “all sites” statement holds only without it.

Richland’s 15.1 K gap between the two paths is slightly wider than the 15 K step from W17 to W32.

Compressor work as extra heat to reject

A chiller moves heat uphill, and the work it does leaves through the same condenser as the heat it moved. The condenser load is the evaporator load plus compressor input: condenser load = evaporator load × (1 + 1/COP).

Efficiency data come from a 2007 Lawrence Berkeley National Laboratory benchmarking study of a corporate data center cooled by 320-ton Trane centrifugal chillers with variable speed drives, rated at 0.55 kW per cooling ton at 100% load.

The part-load ratings are 0.44 kW per ton at 75% load and 0.33 kW per ton at 30%.

One cooling ton is 12,000 Btu/h, or 3.517 kW of heat. The full-load rating of 0.55 kW per ton is then a COP of 3.517 ÷ 0.55 = 6.39, and the condenser multiplier is 1 + 1/6.39 = 1.156. At the 30% rating, 0.33 kW per ton gives a COP of 10.66 and a multiplier of 1.094.

For 1,000 kW of heat on chilled water, the condenser water therefore carries 1,156 kW at the full-load rating and 1,094 kW at the 30% rating. These are rated values for the chillers alone, with no pump or fan power, so they are a lower bound on the heat a chiller plant adds, not a measured result.

With no compressor in the path, the outdoor stage sees only the heat the liquid has already collected. The chiller path hands it 9 to 16 percent more.

Warmer chilled water shrinks that term. Schneider states that each 0.6 °C (1 °F) rise in chilled water temperature raises chiller efficiency 1% to 2%, citing a handbook published in 2000.

The benchmarked plant ran a chilled water setpoint of 42 °F (5.6 °C), so the 9 to 16 percent describes a plant at that setting and not one running warmer.

Schneider frames the trade directly: a shared plant keeps the setpoint inside a narrow band set by the air side, and a separate high-temperature plant frees it. It advises a net present value comparison over ten years or more.

Source: Xu and Greenberg, Data Center Energy Benchmarking Case Study No. 22 (2007), section 2.2.1; Schneider Electric, White Paper 210 (2025), challenge 2.

Old approach values, no hourly weather

The approach values come from a 2011 working group. The paper cites a 1996 ASHRAE handbook page for the tower, a dry cooler vendor’s website for the dry fin cooler, and discussions with IBM and Cray for the server side, and it says the figures are what good design achieves rather than a best case. No current manufacturer selection data was read for this piece, so 5.5 K is a working assumption from a study, not a rating.

Design-hour arithmetic says nothing about annual hours. The sums above cannot say how many hours a dry cooler alone holds a given supply temperature, which needs hourly weather for a site and selection data for a real coil. Fan energy and water spray volume are not quantified here, and every location is in the United States.

Schneider uses 30 °C for the TCS in two ways. One passage reports that key industry players proposed a minimum TCS coolant temperature of 30 °C, and a later one calls 30 °C the current industry maximum. The calculation in the final section only uses 30 °C as the supply the CDU must deliver, which holds under either reading.

The chiller figures describe one plant from 2007 at its rated conditions. None of this settles a design, and equipment sizing and selection need qualified engineering review.

Dry-bulb against wet-bulb at the outdoor boundary

Water meets air at the last coil or fill. A tower cools by evaporation, and the Energy Design Resources guide says about 1% of the flow evaporates for each 12.5 °F (6.9 K) of cooling, while dry-bulb temperature has an insignificant effect on tower performance. A dry cooler has no such dependence on humidity. The water a tower spends is the subject of why water use gets its own metric.

Take a TCS supply of 30 °C and the 4 °C CDU approach in Schneider’s Figure 4. The FWS can then be at most 30 − 4 = 26 °C. A dry cooler at 5.5 K carries the load alone only while dry-bulb stays at or below 26 − 5.5 = 20.5 °C. A tower with its exchanger, 5.6 K in total, needs wet-bulb at or below 26 − 5.6 = 20.4 °C.

Against the paper’s design-hour data, the coolest dry-bulb of the 16 locations is Berkeley’s 27.7 °C, which is 7.2 K above that limit. Only five of the 16 have a design wet-bulb as low as 20.4 °C. At the design hour, no location can run on a dry cooler alone at that limit, and the tower path alone covers only those five.

The second stage is where the boundary charges its price. A spray and a tower pay in water, and a chiller pays in compressor work and returns the 9 to 16 percent extra heat computed above.

Trane’s 2026 white paper adds a sizing point. In a water-cooled chiller plant the dry coolers are typically sized for the chiller condenser load alone. That covers full free cooling or full mechanical cooling but can limit free cooling in hybrid operation, when the economizer and the condenser both need the coil. A coil bought for condenser duty is therefore not automatically sized for hybrid operation.

Source: Trane, Optimizing Data Center Cooling with Economizer and Free Cooling Strategies (2026).

NREL’s high-performance computing data center builds the sequence into one plant. Its IT heat enters one closed loop and is first reused for the building. When temperatures permit, it then goes to a thermosyphon cooler, a dry cooler that uses refrigerant in a passive cycle, and the remainder goes to cooling towers. Over two October days, with dry-bulb between 4.4 °C and 16.7 °C and an IT load of 900 to 950 kW, the thermosyphon cooler rejected the most.

Source: Sickinger et al., NREL/TP-2C00-72196 (2018), sections 1.2 and 2.1.

Server heat crosses a cold plate, a CDU, sometimes a plate exchanger or a chiller, and finally a coil, and the last hop is set by the outdoor reference: dry-bulb for a dry cooler, wet-bulb for a tower. At the studied design hours that difference reaches 16.5 K, more than the step from W17 to W32. With a 30 °C coolant, a dry cooler alone holds only while the air stays at or below 20.5 °C, and whatever covers the remaining hours, whether a water spray, a tower or a compressor, carries the cost of the boundary. How many hours that is depends on hourly weather and a real coil selection, which a qualified engineer has to supply.

References

• “Hot” for Warm Water Cooling / H. Coles (Lawrence Berkeley National Laboratory), M. Ellsworth (IBM), D. Martinez (Sandia National Laboratories) / SC11, November 2011 / https://datacenters.lbl.gov/sites/default/files/Hot%20for%20Warm%20Water%20Cooling.pdf
• Emergence and Expansion of Liquid Cooling in Mainstream Data Centers / ASHRAE Technical Committee 9.9 / 2021 / https://www.ashrae.org/file%20library/technical%20resources/bookstore/emergence-and-expansion-of-liquid-cooling-in-mainstream-data-centers_wp.pdf
• Direct Liquid Cooling System Challenges in Data Centers, White Paper 210 Version 1 / Schneider Electric Data Center Research & Strategy / 2025 / https://media.datacenterdynamics.com/media/documents/WP210_V1_EN.pdf
• Optimizing Data Center Cooling with Economizer and Free Cooling Strategies, DC-WPR006A-EN / Trane / June 30, 2026 / https://elibrary.tranetechnologies.com/public/commercial-hvac/Literature/White%20Paper/DC-WPR006A-EN_06302026.pdf
• Chilled Water Plant Design Guide / Energy Design Resources / December 2009 / https://datacenters.lbl.gov/sites/default/files/EDR_DesignGuidelines_CoolToolsChilledWater.pdf
• Data Center Energy Benchmarking Part 5, Case Studies on a Corporate Data Center (No. 22) / T. Xu and S. Greenberg, Lawrence Berkeley National Laboratory / August 2007 / https://datacenters.lbl.gov/sites/default/files/Benchmarking%20case%20study%2022_2007.pdf
• Thermosyphon Cooler Hybrid System for Water Savings in an Energy-Efficient HPC Data Center: Results from 24 Months and Impact on Water Usage Effectiveness, NREL/TP-2C00-72196 / National Renewable Energy Laboratory / September 2018 / https://datacenters.lbl.gov/sites/default/files/Thermosyphon%20Cooler%20Hybrid%20System%20for%20Water%20Savings%20Paper.pdf