How much energy will a chiller use as a data hall fills?

How do you calculate chiller energy for each buildout year?

Anvilfield estimates about 0.42 GWh in year 1, rising to 1.48 to 1.62 GWh in year 4, for a 275-ton water-cooled duty plant using 2014 OEM points reproduced in GSA/ORNL's revised August 2017 report. Four-year electricity totals 3.738 GWh for the screw option and 3.559 GWh for the magnetic-bearing option, including auxiliaries.

Calculate each year from its load hours and matching package, pump, tower and standby power. For a site forecast, replace the paired rating conditions with the hall's joint load and weather data.

Use the power map, then apply the hours

Multiply each running chiller's electrical power by its operating hours, then sum across chillers and intervals for the year. For cooling measured in tons, multiply tons by kW per ton to obtain electrical power. For cooling measured in thermal kW, divide by the coefficient of performance, which is thermal output divided by electrical input. Keep thermal kW and tons in their matching formulas.

CHW means chilled water; CW means condenser water. Read the power map at the interval’s load, water temperatures, flows and enabled compressors. Annual intensity is total kWh divided by total cooling ton-hours.

Set the boundary at package terminals plus external chilled-water and condenser-water pumps, tower fans and plant controls/standby. CRAH fans, CDUs and UPS losses remain separate. Plant intensity does not establish whole-facility PUE. For air-cooled equipment, use the OEM dry-bulb map, retain package condenser fans inside package power, remove absent tower/condenser-pump rows and add actual glycol or free-cooling circuit loads.

Anvilfield uses four normalized operating years of 8,760 hours each; a calendar year containing February 29 has 8,784 hours. See the HVAC and refrigeration guides for related plant procedures.

Why doesn’t IPLV predict a phased hall’s annual electricity?

Integrated part-load value (IPLV) compares package performance under a prescribed rating profile. It does not contain this hall’s deployment dates, weather, equipment dispatch or external auxiliary electricity. Nonstandard part-load value (NPLV) changes rating conditions without creating the project’s operating schedule.

Daikin CAT 628-3, June 2022, and EnergyPlus 25.1’s discussion of AHRI 550/590-2011 describe legacy coefficients of 1%, 42%, 45% and 12% at 100%, 75%, 50% and 25% load respectively. Those are rating coefficients, not the hall’s occupancy hours. Daikin catalog; EnergyPlus rating calculation.

Keep efficiency units and standard editions straight

For these historical kW/ton points, divide each rating coefficient by its matching efficiency, add the results and take the reciprocal. Both calculated ratings round to 0.33 kW/ton; the screw selection has the lower unrounded value. COP and EER use a different unit form.

Across the four-year schedule, each load bin receives 25% of elapsed hours, while the 25/50/75/100% bins deliver 10/20/30/40% of total ton-hours. Annual kWh per ton-hour therefore uses cooling-output weights. Multiplying IPLV by this schedule’s ton-hours discards the distribution that determines its electricity.

The AHRI catalog checked November 5, 2026 lists AHRI 550/590-2026 alongside retained 2023 material. The numerical rating example here belongs to the dated selections and legacy references.

What should the OEM curve comparison contain?

Compare identical cooling duty and water conditions using Appendix A, printed pages 22, 24 and 25, of GSA/ORNL’s revised August 2017 report. Carrier’s selection is dated November 14, 2014; Daikin’s is dated October 14, 2014, software 09.64.

The 2014 Carrier 23XRV and Daikin WMC400DC selections reproduced in GSA/ORNL's revised August 2017 report rate at about 0.30 and 0.27 kW/ton respectively at 25% of a 275-ton duty. Both use 44°F leaving chilled water, 65°F entering condenser water, 660-gpm evaporator flow and 825-gpm condenser flow.

The screw is lower at 50% load; the magnetic-bearing selection is lower at 25%, 75% and 100%. All points use water and 44°F leaving chilled water.

Ask for low-load points and hot-weather points separately

Request OEM points at 10%, 15%, 20%, 25%, 50%, 75% and 100% selected duty, including low-load operation at high condenser-water temperature. Treat an OEM response of outside operating envelope as an operating boundary, never as a zero-power cell. Specify refrigerant, fluid/concentration, gross/net capacity, fouling, power boundary, redundancy modes and dated digital selection export.

The Daikin selection’s additional page gives 55 tons at 20% load and 19.5 kW, while its 10% point is outside scope. At its stated relief conditions, the 20% power exceeds the calculated 18.22 kW at 25%.

Request compressor-state data to explain such changes; do not force a monotonically declining curve or extrapolate it to 5% duty. Use chiller types and selection for the equipment background, then return to model-specific points.

What if the load falls between published points?

Use the OEM selection engine or a validated performance map first. For a tabulated grid, document interpolation of package kW between adjacent points at matching flows and water temperatures. Keep interpolation inside one compressor state and the approved envelope; obtain additional points across staging or cycling boundaries. Four rating points with changing condenser temperatures cannot define an independent load-and-temperature surface.

What does the published evidence say about magnetic bearings versus screws?

The agency’s final conclusion is similar performance for the two variable-speed chillers tested side by side. GSA GPG-031, updated November 2017 reports averages of about 0.62 kW/ton for the screw and 0.70 kW/ton for the magnetic-bearing machine, with measurement uncertainty spanning 24% less to 4% more screw energy.

The underlying report discusses unequal operating conditions and a dispute about the selected magnetic-bearing frame/compressor sizing. Its field points and fitted curves are different evidence from the OEM selection points in Appendix A.

GSA/PNNL’s November 2012 Pine Bluff assessment compared a new 150-ton magnetic-bearing machine with an older rotary-screw machine and found substantial savings. Equipment generation, size and operating profile explain the different comparison; neither establishes a universal compressor-family winner. The Daikin Navigator CAT 638, August 2025 also describes variable-speed screw technology; a fixed-speed historical comparator cannot stand in for every contemporary screw selection.

How does this extend the progressive-loading research?

Zhang, Li and Wang’s 2024 Building Simulation paper, pages 1693 to 1708 models progressive lifecycle loading, weather, staging and free cooling with TRNSYS 18. Its case includes four 4,200-kW chillers, 13°C chilled-water supply and four waterside economizers.

How do you build year-by-year load bins?

Start with hall-by-hall buildout dates, dated energized IT increments and measured utilization or an owner-approved forecast. Multiply actual IT power by the fraction of its heat reaching this loop, sum those contributions and add non-IT heat reaching the loop.

Use actual power for heat accounting. Leased area, nameplate MW and actual IT MW are different inputs. Metered IT power already contains utilization; do not apply it twice.

Do not multiply IT kW by PUE to obtain evaporator load or add compressor heat to that load. DOE/FEMP’s July 2024 data center design guide separates end uses and cooling-system effects. Track liquid-cooled heat served elsewhere before assigning the remainder to this loop.

The dated schedule in Table B1 favors screw by 9,785 kWh, calculated before rounding. The owner-adopted loop load rises on April 1, 2027; retaining 44°F leaving chilled water, 65°F entering condenser water and the auxiliary ledger below gives 523,349 kWh for screw and 533,133 kWh for magnetic.

Use interval data first and bins second

Anvilfield’s worksheet default is 15-minute intervals: 35,040 records per non-leap year or 35,136 per leap year. Use five-minute event trends during staging tests, with faster event capture where the sequence requires it. Use interval-average kW or meter kWh differences; isolated spot readings miss cycling. Utility billing and contractual requirements govern their own logging.

Begin with joint bins of 5 percentage points load and 5°F entering condenser-water temperature, separated by mechanical, partial and full free-cooling modes. Refine bins at staging changes. Keep timestamps and split transitions; cycling and storage need chronology.

Assign condenser-water temperature from weather and plant operation. A 25% hall in hot weather does not automatically receive 65°F condenser water. Report unserved cooling hours and energy separately, never as low consumption.

Which chillers run while the hall is partly occupied?

The worked plant installs two 275-ton machines, with one serving duty and one stopped but energized in standby. Installed capacity is 550 tons; modeled duty capacity is 275 tons. The adopted standby input is 0.3 kW.

At 68.75 tons, one running 275-ton machine carries 25% of selected duty while the load equals 12.5% of installed capacity. Two running machines would each carry 12.5% at that capacity. Obtain low-load power maps and stable-operation limits before annualizing active load sharing.

Keep selected-duty fraction separate from dispatch part-load ratio. Divide delivered tons by reference duty for selected-duty fraction, and by interval-available capacity for dispatch PLR; they coincide here. A derated available-capacity ratio cannot silently become the coordinate on a reference-capacity performance curve.

Release new equipment on its actual phase date. Apply reliability and operating-envelope constraints before selecting the feasible combination with lowest energy. If required cooling exceeds availability, stage feasible equipment or record unmet cooling.

Use startup and acceptance evidence to establish availability and chiller sequencing and reset to define dispatch. Stopped standby, active sharing and minimum-flow circulation each need their own compressor, pump and heater inputs.

Where do minimum flow and minimum stable load constrain the answer?

Hydraulic minimum flow and refrigeration turndown are separate constraints. Lower thermal demand does not authorize lower evaporator/condenser flow or a faster flow ramp. The Daikin WMC400DC selection dated October 14, 2014 specifies 271.6 gpm minimum evaporator flow against 660 gpm selected design flow, a calculated hydraulic floor of 41.15% that is separate from its minimum stable cooling load. Selection, printed pages 24 to 25.

Calculate bypass and honor the ramp denominator

For water, multiply cooling tons by 24 and divide by the water temperature difference in °F to estimate gpm. At an adopted 10°F load-side water delta-T, 68.75 tons needs about 165 gpm.

One active variable-primary evaporator held at 271.6 gpm needs about 106.6 gpm of supply-to-return bypass at that load. After mixing, evaporator delta-T is approximately 6.08°F.

In primary-secondary and variable-primary pumping, common-pipe flow follows the actual primary and secondary flows. Distinguish expected fixed-flow temperature dilution from low chilled-water delta-T caused by mixing or controls.

Neither later manual retroactively governs the 2014 selection. Confirm the manual and envelope for the supplied machine; a favorable energy result never overrides its flow limits.

Below stable refrigeration capacity, model the permitted cycling, storage, smaller-machine dispatch or false loading. EnergyPlus 25.1 represents cycling and false loading explicitly. Preserve the centrifugal operating envelope and surge control boundary instead of extending a smooth efficiency polynomial below its valid domain.

Worked example: which selection uses less over four buildout years?

Anvilfield estimates that the magnetic-bearing option uses 179,471 kWh less plant electricity over four years under Tables A, B and E. This reverses the published IPLV order, which favors screw.

Anvilfield estimates use the Carrier selection dated November 14, 2014 and Daikin selection dated October 14, 2014 reproduced in GSA/ORNL’s revised August 2017 report, checked November 5, 2026. Multiply exact fractions of 275 tons by published kW/ton and annual hours, then add Table E auxiliaries. The four normalized years retain 44°F leaving CHW, paired entering CW of 65/65/75/85°F at 25/50/75/100% load, and equal auxiliary inputs.

Show one calculation the reader can reproduce

The four-year 6,022,500 ton-hours produce plant intensities of approximately 0.62 and 0.59 kWh per ton-hour, with identical cooling delivered. Round kWh half up, calculating differences and totals before rounding; displayed sums can differ by 1 kWh.

Year 2 favors screw by 6,925 kWh because most hours occur at 50% load. Year 4 favors magnetic by 139,326 kWh and supplies 77.63% of the full-horizon difference.

The first two years have a much smaller decision margin. For years 1 and 2 combined, Anvilfield estimates 1,048,875 kWh for screw and 1,041,626 kWh for magnetic: only 7,249 kWh, or 0.69%, separates them. Show both the early-occupancy margin and the four-year 4.80% result.

How much auxiliary energy belongs in the plant ledger?

Every electrical load inside the declared plant boundary belongs in the ledger, even when cooling demand is small. ASHRAE’s 2024 HVAC Systems and Equipment Handbook, Chapter 3 treats chiller, pump and tower interaction as a plant optimization problem. GSA submetering guidance distinguishes equipment and system measurement; use both to reconcile the boundary.

Both pumps maintain the source flows of 660 gpm chilled water and 825 gpm condenser water. Constant selected flow earns no automatic cubic-law pump saving as thermal load falls. For variable flow, use pump curves, static/differential-pressure requirements, bypass and motor/drive efficiency. DOE’s Improving Pumping System Performance, second edition, May 2006 explains these system-curve constraints.

The adopted year-1 auxiliary electricity exceeds either package’s electricity. Across four years auxiliaries total 991,632 kWh, or 26.52% of screw-plant consumption.

Replace tower estimates using cooling-tower operation and heat-rejection maps. Tower duty includes evaporator heat, compressor heat reaching the condenser and applicable pump heat; equal fan inputs here are a comparison assumption.

Keep inclusion and exclusion visible

Add actual basin heaters, heat tracing, oil/crankcase heaters, standby circulation and glycol-system loads at their correct boundary. Free cooling stops compressor consumption during eligible hours but retains pumping, heat rejection and controls; compressor-off hours are not zero-electricity plant hours.

How do weather, water temperatures and free cooling change the result?

Replace paired rating relief points with joint load/weather intervals for procurement. Use the same deployment dates, weather series, delivered cooling and water setpoints across alternatives.

Resolve water temperatures through the tower, heat exchanger and controls before reading chiller power. Wet bulb governs evaporative heat-rejection opportunity; dry bulb belongs in air-cooled and dry-cooler maps. Optimize total plant kW: colder condenser water reduces compressor work but can require enough extra tower-fan power to erase that saving, as Daikin CAT 628-3 explains.

Raising cooling-water supply from 24°C to 32°C increases calculated power by 28.29% in the PolyU source model at 12°C chilled-water supply. Power rises from 527.37 to 676.55 kW.

Daikin’s WMC-D CAT 628-3, June 2022, page 14 distinguishes 40°F minimum startup entering condenser water from 55°F minimum operating entering condenser water, subject to configuration. Startup permission does not establish a continuous reset limit. Use the modeled machine’s approved envelope.

Count eligible economizer hours at the actual load

The PolyU study finds free-cooling time differing by up to 1,442 hours across part-load ratios within the same location in its modeled systems. Calculate eligibility at each load and ambient condition rather than subtracting that quantity from every project. Waterside economizer operation and data center economizer design require tower/heat-exchanger capacity and approach at the actual duty. Separate partial from full free cooling and retain freeze protection and pumping in both modes.

Which assumptions can reverse the preferred option?

Test occupancy delay, auxiliary differences, package-power error and temperatures against the owner’s required modes. Lower electricity from delivering less cooling is a smaller service quantity, not better efficiency.

An additional continuous 5.12 kW on the magnetic-bearing option eliminates its four-year energy advantage of about 179,471 kWh under Anvilfield's stated schedule. For years 1 and 2 alone, Anvilfield estimates a break-even penalty of just 0.41 continuous kW. Divide the 7,249 kWh early-year advantage by 17,520 hours to obtain that threshold. Resolve standby and pump differences before treating the early-year ranking as decisive.

Report the decision margin and its evidence

The slower-occupancy case replaces year 2 with year 1 while keeping years 3 and 4 unchanged. Delivered cooling falls from 6,022,500 to 5,541,250 ton-hours. Report that denominator beside the recalculated totals. When deployment delay also changes the seasons carrying each load, rerun the joint weather schedule instead of shifting annual percentages alone.

The plus/minus 10% stress test changes magnetic package kWh only and holds its auxiliaries fixed. Identify measured, selected, required and assumed inputs. Test occupancy and weather together where evidence links them, and independently to expose either uncertainty.

What should the energy comparison cost and procurement scope include?

At an adopted flat $0.15/kWh, Anvilfield estimates four-year energy costs of $560,774 for screw and $533,854 for magnetic, a $26,921 difference calculated before rounding. Calculate energy-only cost by multiplying each interval's kWh by its tariff and summing the costs.

Demand charges require the utility’s billing interval, coincident peak, rate and any ratchet. Do not price demand from annual mean kW. Apply time-of-use prices to timestamped energy. For lifecycle selection, compare incremental capital, annual energy/demand cost, maintenance, replacements and residual value over the same study period and owner-selected discount basis.

Simple energy-only payback divides incremental capital by annual energy-cost reduction only when that reduction is positive and reasonably constant. This phased schedule needs dated cash flows. Obtain incremental equipment and maintenance prices rather than inferring a payback from the four-year energy difference.

Turn the model into a bid package

Require a dated model-point export, operating-envelope confirmation, phase equipment-release dates, controls integration, calibrated metering and trend exports. Bidders must identify excluded auxiliaries, enabled-standby policy and annual kWh at common loads and temperatures. Mechanical estimators defining chiller, pump and controls packages find upcoming public work in PorchRocket's Bid Control, which tracks HVAC and refrigeration bids and early signals from board meetings, budgets and capital plans.

The owner supplies deployment assumptions; the OEM supplies selections and operating limits; the plant designer defines topology and feasible dispatch; the controls contractor documents implementation. The owner’s contract sets acceptance, required rating edition, retained formulas and revision responsibility.

What must the field checklist and retained record prove?

Use authorized load-bank steps and production trends to verify sustained low-load operation and phase transitions. Commissioning capacity and power records supply matched observations; a short test does not supply annual weather hours. ORNL’s October 2021 monitoring-based commissioning guidance supports continuing operational review after initial acceptance.

Log synchronized package/plant kW, both flows, matched water temperatures, online machines, operating mode, water setpoints, valve/bypass position and weather. Compare matched loads and temperatures before diagnosing efficiency.

Calculate signed reconciliation bias by subtracting metered kWh from modeled kWh, dividing by metered kWh and multiplying by 100. Report valid, estimated and missing hours plus bin-level residuals. Agree a tolerance from sensor uncertainty, coverage and model purpose; annual agreement alone cannot establish data quality.

Check the cooling denominator at low load

Water-side cooling follows flow times measured temperature difference with actual fluid properties. For an assumed 0.2°F error in delta-T, Anvilfield estimates an 8% error at a 2.5°F delta-T but 2% at 10°F, before flow error. Match and calibrate the temperature pair and independently verify flow before attributing high kW/ton to compressor deterioration.

Field checklist

1. Record phase dates, energized IT increments, utilization and heat routing; reconcile cooling delivered to this loop.
2. Separate installed, available, duty and running capacity; retain the denominator used by each OEM curve.
3. Obtain dated OEM exports at matching temperatures, flows, fouling and power boundaries, including low-load/hot-weather points.
4. Enter minimum flows, ramp limits and permitted cycling/storage strategy; flag every interval outside the approved envelope.
5. List each pump, tower, heater, control and standby load once; meter or explicitly estimate its power.
6. Synchronize load, weather, water setpoints and operating mode; preserve partial and full free-cooling intervals.
7. Reconcile valid, missing and estimated hours to the calendar; document leap years and daylight-saving treatment.
8. Verify sustained low-load operation and phase transitions with authorized test steps, calibrated sensors and synchronized trends.
9. Apply energy tariffs and demand billing rules; rerun occupancy, temperature, standby and package-power sensitivities.
10. Save formulas, source maps, CSV exports and calibration records; reconcile by operating bin and update at each material phase or controls change.

Retain a record another engineer can recalculate

Save timestamps, timezone, duration, phase/date, IT kW, cooling tons, active IDs and selected capacities. Retain component kW; chilled-water supply/return and condenser-water entering/leaving temperatures; both flows; wet/dry bulb; mode; bypass position; availability; quality flags; map revision and tariff period. A missing interval requires a flag and an estimate method, not an unmarked zero.

Separate workbook inputs, dispatch, energy/cost calculations and meter reconciliation. Preserve original data, formulas, units, rejected intervals and revisions. Review each load milestone, equipment addition, standby-policy or setpoint change, plus annual meter reconciliation. Deliver CSV exports so another engineer can reproduce the comparison.

Sources and citation

Sources checked and review date: November 5, 2026. Sixteen primary references, including the full PolyU paper, support the comparison.

1. GSA/ORNL, Variable-Speed Screw Chiller, Sidney Yates Building, ORNL/SR-2017/349, revised August 2017. OEM Appendix A; footers retain July 2017.
2. Zhang, Li and Wang, Building Simulation 17 (2024), 1693 to 1708, DOI 10.1007/s12273-024-1167-9. Progressive loading and Table 5 temperature inputs. Full paper.
3. GSA GPG-031, Variable-Speed Direct-Drive Screw Chiller, updated November 2017. Final findings and field uncertainty.
4. ASHRAE, 2024 HVAC Systems and Equipment Handbook, Chapter 3, Central Cooling and Heating Plants. Plant interactions.
5. AHRI, 550/590 and 551/591 catalog, checked November 5, 2026. Catalog listing for AHRI 550/590-2026.
6. Trane, CHS-PRB009J-EN, February 2024, Table 151. RSCW0430F2/RTHD rental-unit boundaries.
7. Daikin, Magnitude WMC D Vintage, CAT 628-3, June 2022. Legacy rating coefficients and water-temperature limits.
8. Daikin, WMC-D IOM 1210-8, January 2023, printed page 9. Current-flow ramp denominator.
9. Daikin, Navigator WWV/TWV IOM 1264-5, August 2026, page 10. Design-flow ramp denominator.
10. GSA, Submetering, live guidance checked November 5, 2026. Equipment and system metering boundaries.
11. EnergyPlus, Engineering Reference 25.1, Chillers, hosted by Big Ladder. Power maps, legacy rating calculation, cycling and false loading.
12. DOE/FEMP, Best Practices Guide for Energy-Efficient Data Center Design, revised July 2024, DOE/GO-102024-6283. End uses and economizers.
13. GSA/PNNL, Variable-speed Oil-free Centrifugal Chiller with Magnetic Bearings Assessment, November 2012. Pine Bluff comparator scope.
14. ORNL, Monitoring Based Commissioning: Step-by-Step Approach to Implementation in Chiller Plants, October 2021. Continuing monitoring workflow.
15. DOE, Improving Pumping System Performance: A Sourcebook for Industry, second edition, May 2006. Pump and system curves.
16. Daikin, Navigator WWV/TWV CAT 638, August 2025. Contemporary variable-speed screw context.

How to cite this page: Anvilfield, How much energy will a chiller use as a data hall fills?, 2026, https://anvilfield.com/field-guides/hvac/chiller-phased-load-energy-comparison/

Related tools

FAQ

People also ask

• What is a factory authorized chiller startup?
• What is chiller plant sequencing?
• How is a data center built?
• What is primary-secondary chilled water pumping?
• What is a waterside economizer?