HVAC
Chilled water pumping: primary-secondary vs variable primary flow
Pick how the water moves through the chillers and out to the coils: constant flow, primary-secondary with a decoupler, or variable primary flow with a minimum-flow bypass, and hold the delta-T.
Direct answer
Chilled water pumping configuration sets how water moves through the chillers and out to the coils. Primary-secondary runs a constant-flow loop through the chillers and a variable-flow loop to the coils, joined by a decoupler. Variable primary flow uses one variable-speed pump set through both, with a minimum-flow bypass protecting the chiller.
Key takeaways
- Primary-secondary pumping runs constant flow through chillers and variable flow to coils, joined by a decoupler; variable primary flow uses one variable-speed pump set with a minimum-flow bypass.
- The decoupler should normally carry a small flow from primary supply to primary return; reverse (deficit) flow signals staging on another chiller and its pump.
- Size the decoupler short and fat: under about 1.5 ft of head friction loss, length 3 to 10 pipe diameters, with no valves or fittings.
- Chiller minimum and maximum evaporator flow comes from the manufacturer's data sheet, not a rule of thumb; below minimum the tubes can freeze and split.
- Variable-flow plants use two-way throttling valves at coils; three-way diverting valves cause low delta-T and waste pump energy, and DP sensors belong out at the far load.
The pumping configuration, and what it decides
The pumping configuration decides how water moves through the chillers and out to the coils, and that one decision drives the plant's pump energy, the chiller's protection, and the complexity of the controls you will live with for the life of the building. Get it right and the plant tracks the load quietly. Get it wrong and you fight low delta-T, tripped chillers, and a pump bill nobody budgeted for.
There are three configurations you meet in the field: constant flow, primary-secondary, and variable primary flow. They differ in one thing, how much the flow is allowed to vary and where. The chiller wants steady flow through its evaporator to protect the tubes. The building wants flow to rise and fall with the cooling load so the pumps are not pushing water nobody needs. The configuration is how the design reconciles those two wants.
Bringing the plant online and proving the chiller is a separate job from picking the scheme, and so is setting the flow at every coil. Those are covered in the chiller plant startup guide and the hydronic balancing guide. This guide is about the scheme itself: how the water moves, why, and what it costs.
Constant flow, the old way, and why it wastes pump energy
Constant-flow systems pump the same gallons per minute through the whole loop all the time, load or no load. A single set of pumps pushes water through the chillers and out to the coils, and the coils use three-way valves that divert water around the coil instead of throttling it back. When a zone needs less cooling, the three-way valve sends the water through a bypass port instead of through the coil, so the coil gets less but the loop flow never changes.
That is the problem. The pump runs at full flow and full power on the mildest day of the year, because the bypassed water still has to be pumped. You are paying to move water that does no cooling. On an old building you can hear it, the pumps running flat out in October.
Constant flow protects the chiller well, because the evaporator always sees its design flow. It is simple, and it still has a place on small plants and some process loads where steady flow is worth the energy. But the penalty is real, and modern energy codes push hard toward variable flow for anything of size. The three-way valve at the coils is the tell that you are looking at the old scheme.
What is primary-secondary chilled water pumping?
Primary-secondary pumping splits the plant into two loops joined by a short pipe called the decoupler. The primary loop runs constant flow through the chillers, with one dedicated primary pump per chiller, so each evaporator always sees the steady flow it needs. The secondary loop runs variable flow out to the coils, with variable-speed secondary pumps that ramp up and down with the building load. The chiller sees constant flow. The building varies. The decoupler lets the two loops run at different flows without fighting each other.
This was the standard answer for decades, and it still shows up on most existing central plants. It protects the chiller by design, because the primary pump and the chiller flow are locked together and the secondary side cannot starve the evaporator. The cost is a second set of pumps and the energy to run the primary loop at constant flow whenever a chiller is on.
The decoupler, also called the common pipe or the bridge, is the part that makes the scheme work and the part most often gotten wrong. How the chiller itself is protected, started by the factory tech, and staged is its own task at startup, handled separately from the pumping design.
Which way should the decoupler flow?
The decoupler should normally carry a small flow from the primary supply to the primary return, which means the chillers are making slightly more flow than the building is using. Read the decoupler and you read the plant. When primary flow is greater than secondary, the surplus chilled water bypasses from supply to return through the common pipe, and the building is getting all the cold water it wants. That is the healthy direction.
When the flow reverses and return water is pulled across the decoupler into the secondary supply, you are in deficit. The building wants more flow than the running chillers are producing, so warm return water mixes into the supply going out to the coils, and the supply temperature to the building creeps up. Deficit is the signal to stage on another chiller and its primary pump.
The thing that matters is direction, not just magnitude. A flow meter in the common pipe, or a pair of temperature sensors across it, tells you which way it runs. The decoupler itself has to have almost no pressure drop, so it is sized short and fat. A common rule keeps the friction loss under about 1.5 ft of head and the length at three to ten pipe diameters, with no valves or fittings in it. Put a balancing valve or a long run of elbows in the common pipe and you have coupled the loops you were trying to decouple.
What is variable primary flow (VPF)?
Variable primary flow drops the second set of pumps. One set of variable-speed pumps pushes water through the chillers and out to the coils, and the flow varies through everything, including the evaporator. There is no decoupler and no secondary loop. The pumps ramp down as the coil valves throttle, so the plant moves only the water the building needs, and the pump energy follows the load down.
That is the appeal: fewer pumps, less pump energy, a smaller mechanical room. Energy codes and modern designs lean toward VPF for that reason. The catch is that the chiller now sees varying flow through its evaporator, and below a certain flow the chiller is in trouble. So VPF needs a control the constant-primary schemes never did.
The piece that makes VPF safe is the minimum-flow bypass. A modulating bypass valve sits near the chillers and opens when the building demand falls below the chiller's minimum evaporator flow, recirculating just enough water to keep the running chiller above its limit. VPF also has to manage how fast the flow changes and how chillers stage, because adding or dropping a chiller on a shared variable loop shifts flow through the others. It is the lower-energy scheme, but it asks more of the controls and the commissioning.
The chiller minimum flow, and what protects the evaporator
The chiller's evaporator has a minimum flow below which it will not run safely, and that number comes from the manufacturer, not from a rule of thumb. Below the minimum, water velocity in the evaporator tubes drops far enough that the flow goes laminar, heat transfer falls off fast, and the tubes no longer pull heat out evenly. At the extreme, the water near the tube wall gets cold enough to freeze, and a frozen evaporator splits tubes and ends the chiller. This is the failure VPF has to design around.
The manufacturer publishes both a minimum and a maximum evaporator flow, usually as a tube velocity. A common figure is roughly 1.5 ft per second minimum tube velocity at half load or lower, and 3 to 12 ft per second across the full operating range, but the machine's data sheet is the only number that counts. Two chillers of the same tonnage from different makers can have different minimums.
The minimum-flow bypass valve in a VPF plant exists to hold the evaporator above that limit. The chiller controls also limit how fast flow is allowed to change, because a sudden swing can trip the machine on a flow fault or upset the refrigerant control. In primary-secondary, the dedicated primary pump handles this by keeping flow constant. In VPF, the bypass and the pump control are what keep the chiller alive, so the manufacturer's minimum flow and rate-of-change limits govern the whole design.
What is low delta-T syndrome?
Low delta-T syndrome is when the chilled water comes back to the plant colder than it should, so the temperature difference across the coils is smaller than design. If the system was drawn for a 12 to 16 degree F rise and the plant only sees 6 or 8, you have it. The plant compensates the only way it can, by pumping more water to move the same heat, and on a primary-secondary plant the secondary flow climbs until it exceeds the primary and the decoupler runs backward into deficit. Now you are staging chillers to make flow, not to make cooling, and the kW per ton goes to ruin.
It is the most common chronic problem on chilled water plants, and it is rarely the plant's fault. The cause is out at the coils and valves. Three-way valves blend cold water straight to the return. Control valves with poor authority hunt and average out to partly bypassed. Fouled coils cannot transfer heat, so the water leaves nearly as cold as it came. Oversized coils at part load hit setpoint with barely any rise. Each of these returns cold water to the plant.
The fix lives at the coil, not the pump. Chasing low delta-T means checking coil cleanliness, valve authority and selection, and the control sequence, which is balancing and commissioning work. A plant cannot pump its way out of a coil problem, though plenty of operators try, and the higher flow just makes the pump bill worse while the delta-T stays flat.
The pump VFD and differential-pressure control
The variable-speed pumps ride on a differential-pressure (DP) setpoint. A sensor reads the pressure difference between supply and return, usually out near the most remote or most demanding load, and the drive speeds the pump up or slows it down to hold that DP as the coil valves open and close. Open a valve and the pressure sags, so the pump speeds up. Close valves and the pressure rises, so the pump backs off. That is the loop that turns coil demand into pump speed.
Where the DP sensor sits matters more than people expect. Put it at the pump and you hold full head all the time and lose most of the savings. Put it out at the far coil and the pump only makes the head the system actually needs. The setpoint itself should be the lowest DP that still satisfies the worst-case coil at design flow, which is a number you find during balancing, not one you guess.
The better plants go a step further with DP reset. Instead of holding one fixed setpoint, the control watches the coil valve positions and lowers the DP setpoint until the most-open valve is nearly wide open. That keeps every valve working and drops the pump speed as low as the load allows. Setting and proving the DP setpoint and reset is part of the water-side balancing and commissioning, where the design intent meets the actual system curve.
Variable speed and the affinity laws
The reason variable flow saves so much is the affinity laws. Pump flow tracks speed, head tracks the square of speed, and power tracks the cube of speed. Slow a pump to half speed and it moves half the water at a quarter of the head for one eighth of the power, in the ideal case. That cube relationship is the whole argument for variable flow.
The savings live at part load, which is where the plant spends almost all its hours. A chilled water plant rarely runs at design load. It runs at 40 to 70 percent most of the time. At 70 percent flow the pump draws roughly a third of full power, ideally. Real pumps fall short of the textbook cube because motor and drive efficiency drop at low speed and because the system has fixed head the pump still has to overcome, but the part-load savings are large and real, and they compound over thousands of run hours.
The fixed head is the catch worth naming. The cube law applies to the friction head that varies with flow, not to the static lift or the minimum DP the system has to hold. The more of your total head is fixed, the flatter the savings curve. This is why holding the DP setpoint low, and resetting it lower when the loads allow, matters as much as the variable drive itself.
Two-way valves at the coil, not three-way
Variable-flow systems use two-way control valves at the coils, and that choice is what lets the flow vary at all. A two-way valve throttles. When the coil needs less cooling, the valve closes down, less water flows, the loop flow drops, and the pump can slow. A three-way valve diverts. It sends water around the coil through a bypass port, so the coil gets less but the loop flow stays the same. Three-way valves are why constant-flow systems waste pump energy, and they are a direct cause of low delta-T, because the bypassed water blends cold back into the return.
So the rule on a variable-flow plant is two-way valves at the coils and no three-way bypasses scattered through the system. Every three-way valve or open balancing bypass left in a variable-flow loop is a path for cold water to short-circuit back to the plant, which flattens the delta-T and props up the flow the pump is trying to reduce.
The one deliberate bypass that belongs in a VPF plant is the single minimum-flow bypass at the chillers, and that one is controlled, not a fixed open port. It opens only when the building demand falls below the chiller minimum. Everything else should be two-way and should be allowed to actually throttle.
Staging chillers and pumps to the load
Staging is matching the number of running chillers and pumps to the cooling load, and the pumping configuration changes how you do it. The plant should run the fewest machines that can carry the load at good efficiency, then add and drop them as the load moves. Stage on too late and you go into deficit and lose supply temperature. Stage on too early and you run extra equipment at poor part-load efficiency.
In primary-secondary, the decoupler tells you when to act. Deficit flow, return water crossing into the secondary supply, means the running chillers cannot make the flow the building wants, so you stage on the next chiller and its dedicated primary pump together. Surplus tells you a chiller may be able to come off.
In variable primary flow, staging is harder, because all the chillers share one variable loop and adding a chiller redistributes flow through the others. The control has to bring on the next chiller and adjust pump speed without dropping any running machine below its minimum flow or spiking the others, and it has to manage the bypass through the transition. This is where VPF earns its reputation for needing good controls. How the chillers themselves are protected and proven through these transitions at startup is its own commissioning task.
Data center chilled water and N+1 pumping
Data centers change the priorities. The cooling load runs around the clock and barely follows the weather, the supply temperature has to stay tight to hold the rack inlet conditions, and a loss of cooling is a loss of uptime measured in minutes. So the pumping is built for redundancy first and efficiency second, though both matter.
Redundancy shows up as N+1 or 2N pump and chiller counts, one more pump and chiller than the load needs, so a single failure or a maintenance outage never drops the cooling. The pumps are often headered so any pump can serve any chiller, and the controls have to fail over without a flow interruption the racks would feel. Concurrent maintainability, the ability to service a pump or chiller while the plant keeps cooling, is the design goal on the higher tiers.
Variable flow still pays here because of the run hours, but the minimum-flow protection and the staging logic carry more weight, because the plant cannot be allowed to trip a chiller while chasing efficiency. Many data center plants also push higher chilled water temperatures and a wider design delta-T to gain economizer hours and pump less water, which makes the low delta-T discipline at the coils worth even more.
Primary-secondary vs variable primary flow: which should you use?
Variable primary flow generally uses less pump energy and less plant space than primary-secondary, because it drops a whole set of pumps and lets the chiller flow ride down with the load. That is why new designs lean toward it. Primary-secondary protects the chiller more simply, with a dedicated constant-flow primary pump per machine, and its staging logic off the decoupler is easier to get right, which is why it dominates the existing building stock and still gets specified where simplicity or an older chiller's flow limits argue for it.
The honest answer is that the project decides, on the chillers' flow tolerance, the load profile, the controls the operating staff can actually run, and the energy code. A chiller that tolerates variable flow well and a control contractor who can do the staging make VPF the lower-cost-to-run choice. An older chiller, a process load that wants steady flow, or a plant that will be run by a small staff can make primary-secondary the safer call. Neither is wrong. The mistake is picking the scheme by habit instead of by the chiller data and the load.
| Factor | Primary-secondary | Variable primary flow |
|---|---|---|
| Pumps | Primary pump per chiller plus secondary pumps | One variable-speed pump set, no secondary |
| Pump energy | Higher, primary loop runs constant | Lower, flow rides down with the load |
| Chiller protection | Simple, dedicated constant primary flow | Needs minimum-flow bypass and flow-rate control |
| Staging | Off the decoupler flow direction | Shared loop, harder, controls-dependent |
| Decoupler | Required common pipe | None |
| Best fit | Existing plants, older chillers, steady-flow loads | New plants, capable controls, variable load |
Commissioning the pumping configuration
Commissioning the pumping is proving the water moves the way the design intended across the whole load range, not just at design day. The checks are specific to the scheme. On primary-secondary, you verify the decoupler flows the right direction at several load points and goes into deficit only when it should, that the primary pumps and chillers stage together, and that the secondary DP setpoint holds the far coil at design flow. On variable primary flow, you prove the minimum-flow bypass opens before any chiller drops below its minimum evaporator flow, that the pumps hold DP, and that staging a chiller up or down never starves a running machine.
Across both, you verify the DP reset works if it is part of the sequence, and you confirm the delta-T at the coils and at the plant meets design at part load, because that is where low delta-T hides. A plant that makes design delta-T at full load can fall apart at 40 percent, and 40 percent is where it lives.
This work overlaps the chiller startup and the water-side balancing, and it should be sequenced with them. The chiller proven and started by the factory tech, the coils balanced to design flow, then the pumping and staging proven across the load range. Skip the part-load points and you have commissioned the one condition the plant rarely sees.
Keeping it running after turnover
The owner inherits three things that drift, and all three quietly cost energy if nobody watches them. The minimum-flow bypass valve in a VPF plant has to keep modulating correctly. A bypass that sticks open recirculates water all the time and floods the plant with low delta-T, and a bypass that sticks shut risks the chiller on a low-load morning. It is a valve and an actuator, and like any valve it needs checking.
The DP setpoint is the second. It gets bumped up over the years to chase a complaint about a warm zone at the far end, and it never gets put back, so the pumps run harder than they need to for the life of the building. Every increase should be questioned, because the real fix for that warm zone is usually a balancing or coil problem, not more pump head.
The delta-T is the gauge for all of it. An operator who trends the plant delta-T sees low delta-T syndrome coming before it shows up on the energy bill, and sees a stuck bypass or a failed valve as a step change in the trend. The plant that holds its design delta-T at part load is the plant that was commissioned right and is being watched. The one that has quietly slid to a 6 degree rise is paying for it every hour.
What to document
The record that matters is the one that says how each loop is supposed to behave, so the next person can tell drift from design. For each loop, write down the configuration, the pumps and their drives, the minimum flow the chiller requires, the DP setpoint and where the sensor lives, and the design delta-T. Without that, an operator three years in has no way to know whether the plant is running the way it was built or has wandered off.
| What to record | Detail to capture |
|---|---|
| Loop and configuration | Constant flow, primary-secondary, or variable primary flow |
| Pumps and VFDs | Pump count, duty and standby, drive and speed control |
| Chiller minimum flow | Minimum and maximum evaporator flow, from the data sheet |
| Minimum-flow bypass | Valve, control point, and the flow at which it opens |
| DP setpoint | Value, sensor location, and the reset sequence if any |
| Design delta-T | Supply and return temperatures and the design rise |
| Decoupler | Size and normal flow direction (primary-secondary) |
Common mistakes
- Ignoring low delta-T and pumping more water to compensate instead of fixing the coils, valves, and controls that cause it.
- Running a variable primary flow plant with no minimum-flow bypass, or one sized smaller than the chiller's minimum evaporator flow.
- Letting the decoupler run backward into deficit at part load when the plant should be staging on another chiller.
- Leaving three-way valves or open balancing bypasses in a variable-flow loop, which flattens delta-T and props up the flow.
- Holding the pump at a fixed DP setpoint with no reset, or putting the DP sensor at the pump instead of out at the far load.
- Not staging chillers and pumps to the load, so the plant runs more machines than the load needs at poor part-load efficiency.
- Treating one chiller's minimum flow as a generic number instead of pulling the manufacturer's data sheet for the actual machine.
Field checklist
Want this checklist to run itself on every job — with photo proof and a signed record crews can hand the customer? That's FieldOS.
Standards and references
ASHRAE is where the design framework lives. The ASHRAE handbooks cover chilled water distribution and the primary-secondary and variable-primary schemes, and the ASHRAE energy standard, 90.1, pushes variable-speed pumping and pump-power limits for systems above a size threshold. The exact thresholds and requirements shift between editions, so confirm them against the edition the project and the jurisdiction have adopted.
The number that governs the chiller protection is not in a code at all. It is the chiller manufacturer's published minimum and maximum evaporator flow and the allowed rate of flow change, and that data sheet wins over any rule of thumb. Design a variable primary flow plant without the manufacturer's minimum flow in hand and you are guessing at the one number that can destroy the machine.
The energy code, often based on ASHRAE 90.1 or the IECC depending on the jurisdiction, is what pushes variable-speed pumping and DP reset on plants of size, and the adopted edition and local amendments control. Above all of it sits the project specification and the design engineer's sequence of operations, which set the actual setpoints, staging, and control intent for the specific plant. Cite the document that controls the point, let the manufacturer's data govern the chiller, and let the spec govern the sequence.
Units, terms, and synonyms
Chilled water pumping carries its own vocabulary, and the same idea reads differently across a drawing set, a controls submittal, and a chiller cut sheet. Flow is in gallons per minute, GPM, in US practice and in liters per second or cubic meters per hour in metric. Delta-T is the supply-to-return temperature difference, in degrees F or C. Differential pressure shows up as DP, in feet of head or psi, and pump head is in feet of water.
The configurations go by several names. Variable primary flow is VPF. Primary-secondary is sometimes called a decoupled system or primary-secondary-variable. The decoupler is also the common pipe or the bridge. Knowing the synonyms keeps you from arguing about two names for the same pipe.
- Delta-T
- The supply-to-return chilled water temperature difference, the design rise across the coils
- Decoupler / common pipe
- The short low-resistance pipe joining the primary and secondary loops in a primary-secondary plant
- VPF
- Variable primary flow, one variable-speed pump set serving both the chillers and the coils
- DP setpoint
- The differential pressure the variable-speed pumps hold, ideally sensed at the far load
- Minimum flow
- The lowest evaporator flow the chiller tolerates, set by the manufacturer's data sheet
- GPM
- Gallons per minute, the flow unit in US chilled water practice
- Affinity laws
- Pump flow tracks speed, head the square of speed, and power the cube of speed
FAQ
What is primary-secondary chilled water pumping?
Primary-secondary pumping splits the plant into a constant-flow primary loop through the chillers, with one pump per chiller, and a variable-flow secondary loop out to the coils. A short decoupler pipe joins them, so the chillers see steady flow while the building flow varies with the cooling load.
What is variable primary flow?
Variable primary flow, or VPF, uses one set of variable-speed pumps to push water through the chillers and the coils, with no secondary loop. The flow varies through the evaporator, so a modulating minimum-flow bypass valve near the chillers opens at low load to keep each running chiller above its minimum flow.
What is a decoupler in a chilled water system?
A decoupler, also called the common pipe or bridge, is a short low-resistance pipe joining the primary and secondary loops in a primary-secondary plant. It lets the two loops run at different flows without fighting each other. It must have almost no pressure drop, so it is sized short and fat with no fittings.
What is low delta-T syndrome?
Low delta-T syndrome is when chilled water returns colder than design, so the temperature rise across the coils is too small. The plant pumps more water to move the same heat, the kW per ton climbs, and the secondary flow can exceed the primary. The cause is at the coils and valves, not the plant.
Primary-secondary vs variable primary flow: which is more efficient?
Variable primary flow usually uses less pump energy, because it drops the secondary pumps and lets the chiller flow ride down with the load. Primary-secondary runs its primary loop at constant flow, which costs more. VPF needs better controls and a minimum-flow bypass to protect the chiller, so the project and the chiller data decide.
Why does a variable primary flow system need a minimum-flow bypass?
Because the flow varies through the evaporator, and below the chiller's minimum the tube flow goes laminar, heat transfer collapses, and the evaporator can freeze and split tubes. The bypass valve opens when building demand falls below the chiller minimum, recirculating enough water to hold the running chiller above its limit.
Which way should the decoupler flow?
Normally from primary supply to primary return, meaning the chillers make slightly more flow than the building uses, a small healthy surplus. When return water reverses into the secondary supply, the plant is in deficit and the supply temperature creeps up, which is the signal to stage on another chiller and its pump.
Why are two-way valves used instead of three-way valves?
Two-way valves throttle, so when a coil needs less cooling the flow drops and the pump can slow, which is how variable flow saves energy. Three-way valves divert water around the coil instead, keeping loop flow constant and blending cold water back to the return, which wastes pump energy and causes low delta-T.
What do I do if my chilled water delta-T is low?
Look at the coils and valves, not the pumps. Check for fouled coils, three-way valves or open bypasses blending cold water to the return, control valves with poor authority that hunt, and oversized coils at part load. Fixing those is balancing and commissioning work. Pumping more water only makes the energy bill worse.
How do I set the pump differential-pressure setpoint?
Set it to the lowest DP that still satisfies the worst-case coil at design flow, with the sensor out near that far load, not at the pump. Find the value during balancing. Better still, add DP reset that lowers the setpoint until the most-open coil valve is nearly wide open, dropping pump speed as the load allows.
People also ask
Codes cited in this guide
This guide is written and reviewed against the published standards below. Always confirm the current adopted edition with the authority having jurisdiction.