Rear-door heat exchangers (RDHx) are often described with a single number—“kW per rack.” That’s convenient for a slide deck, but it’s not how RDHx behaves in the aisle.
In practice, rear door heat exchanger capacity is a curve, not a point estimate. The capacity you can count on depends on:
Inlet water temperature (and your dew point margin)
Water flow rate per door
Airflow through the rack and door (and how much bypass/recirculation exists)
Load volatility (steady enterprise compute vs. AI/HPC spikes)
This FAQ answers the question a data center project director actually needs to resolve:
How much rack heat can an RDHx remove safely and consistently—under my water temperatures, flow constraints, and operational risks?
1) How much heat can a rear door heat exchanger remove (kW per rack)?
A defensible planning answer is a range, not a single number. Many commercial RDHx products are specified in the tens of kW per rack, and some industry overviews describe RDHx as supporting racks roughly in the 20–80 kW band depending on design and operating conditions.
If you need a concrete example of how rated capacity bands are published, HPE’s liquid-cooled door QuickSpecs lists multiple product families with capacity ranges such as 14–35 kW, 35–55 kW, and 55–75 kW, and it also states a key constraint: water temperature should stay 2°C (4°F) above the maximum anticipated dew point (a direct link between capacity and condensation risk) in HPE’s Motivair liquid-cooled door QuickSpecs (2026).
For a practical “hybrid zone” planning band, Coolnetpower’s fit guidance positions RDHx most often around ~20–40 kW per rack as a common application range, with clear caveats that water temperature must stay above dew point and that performance depends on rack conditions in Coolnetpower’s chilled-water RDHx fit guide for 20–40 kW racks (2026).
2) Why do RDHx “kW per rack” numbers vary so much?
Because RDHx capacity is condition-dependent. Two sites can buy “the same class” of RDHx doors and see different outcomes if they differ in:
dew point control (how cold the water can safely be)
available flow (pump and distribution limits)
containment and sealing discipline
door control strategy (passive vs. active)
A useful way to de-risk procurement is to stop asking for “kW per rack” in isolation and start asking for capacity at your water temperature band and flow.
What ratings assume (capacity curves, approach temperature, and ΔT)
3) What does an RDHx capacity curve usually show?
Most RDHx capacity curves express heat removal (kW) as a function of operating conditions. You’ll typically see:
Capacity (kW) on the y-axis
and one or more of:
inlet water temperature (°C/°F)
water flow rate (GPM/LPM)
sometimes airflow (CFM) or fan speed (for active doors)
How to read the curve for your case (you chose 16–20°C / 61–68°F chilled-water supply as the example band):
Select your inlet water temperature (16°C vs 20°C can be a meaningful change).
Select your flow constraint per door (what you can actually deliver during peak).
Read steady-state capacity, then apply a realism discount for bypass air, uneven racks, fouling risk, and transients.
4) What “approach temperature” to water is assumed in ratings?
For RDHx, “approach” is best treated as a practical effectiveness limit: how close the cooled exhaust (leaving air condition) can get to the water temperature under load.
Smaller approach generally requires more heat exchanger effectiveness and/or higher airflow/flow.
Larger approach means the door can’t pull the air condition as close to water temperature, lowering effective capacity at a given inlet water temperature.
The procurement-critical link is that approach is constrained by condensation control. Specifications often require a dew point margin (for example, the 2°C / 4°F margin in the HPE QuickSpecs). That margin caps how cold you can run—especially during economizer seasons or humidity excursions—which reshapes the entire capacity curve.
5) How does performance vary with water ΔT and flow rate?
On the water side, the physics is straightforward:
(Q = dot m cdot C_p cdot Delta T)
Where (Q) is the heat transferred to water, (dot m) is coolant mass flow, (C_p) is water’s specific heat, and (Delta T) is the water temperature rise across the door.
A worked example that’s directly relevant to RDHx measurement uses 20 gpm and a 3.8°C water temperature rise to compute 20 kW transferred in Electronics Cooling’s heat-to-water calculation example (2011).
What this means in practice:
If you’re flow-limited, you may not be able to hold capacity during spikes.
If you’re trying to keep ΔT “too small” (often due to control choices), you can end up pushing unnecessary flow to achieve the same kW.
Important: water-side math does not automatically equal “rack heat removed.” Bypass air and recirculation can cause you to overestimate effective capture unless you measure carefully.
6) What inlet/outlet air temperatures are typical with RDHx?
Air temperatures are best treated as commissioning outcomes, not brochure numbers. That said, practical guidance can help anchor expectations.
Coolnetpower’s fit guide notes that server outlet air temperatures can be reduced by about 10°F to 35°F depending on flow, coolant temperature, and rack conditions (again: this is behavior on a curve, not a constant) in the same 20–40 kW planning context.
Key Takeaway: The fastest way to lose performance is to measure “one convenient point” and assume it represents the rack. Use multiple sensors (rack exhaust, door discharge, hot aisle) and correlate with heat-to-water measurements.
Behavior under AI/HPC spikes (and what “safe” really means)
7) What happens during transient load spikes?
During a spike, the bottleneck is usually response—not steady-state capacity. Load can step up in seconds. Your ability to keep temperatures stable depends on:
whether airflow through the door can increase fast enough (server fans and/or door fans)
whether valves and pumps can deliver the required flow at the right time
sensor placement and control loop tuning
the thermal inertia of the coil, water loop, and rack airflow path
A practical planning stance for “safe and consistent” performance:
Define three rack load numbers: sustained, 95th percentile, and peak.
Require performance evidence at your water temperature band (16–20°C here) and your expected flow range.
Assume some portion of short spikes will pass through to the room unless proven otherwise.
8) Passive vs. active RDHx: what changes under spikes?
Passive RDHx relies on server fan curves. Under spikes, fans ramp based on inlet temperature and internal controls; door airflow is “whatever the servers provide.”
Active RDHx adds fans at the door, which can provide more predictable airflow through the coil at higher loads—but introduces another control loop and another operational dependency.
If your racks are volatile (common in AI clusters), active control can improve stability if you have a clear control philosophy, dew point override behavior, and alarm handling.
Limits you should assume (uneven racks, blanking, and containment)
9) Any limits for uneven rack population and missing blanking?
Yes—and these are often the difference between “works on paper” and “stable in production.”
Uneven racks and missing blanking panels reduce effective RDHx capacity because they change airflow paths:
Bypass air short-circuits the intended flow path through the heat exchanger.
Recirculation puts warm air back into inlets, driving fan ramps and destabilizing temperatures.
Localized heat flux (e.g., GPUs concentrated in one zone) creates hot spots even when average rack kW looks acceptable.
If you want performance to match the curve you’re buying, treat airflow management as part of the system scope.
10) What commissioning checks correlate best with stable capacity?
A short procurement-friendly checklist:
dew point control strategy and fail-safe behavior
verified flow per door (not just valve position)
heat-to-water metering plan (flow + ΔT + sensor tolerances)
airflow integrity (blanking panels, sealed openings, containment)
alarm rationalization (actionable thresholds, not noise)
⚠️ Warning: If your water temperature can drift to (or below) room dew point during economizer seasons, you need a defined control strategy (dew point override, mixing/secondary loop, and alarm behavior). Treat this as a design requirement.
Why this matters for energy (and how to talk about it credibly)
11) Does PUE improve automatically if RDHx removes more heat to water?
Not automatically—but it can improve when the system is integrated correctly. RDHx can shift heat rejection away from room air systems, enabling changes like higher chilled-water setpoints, lower CRAH/CRAC fan energy, and more economizer hours (climate-dependent).
Coolnetpower covers the conditions and caveats in Coolnetpower’s RDHx PUE FAQ (2026).
As first-party context, Coolnetpower has also shared a scenario comparison table indicating that moving from a traditional air-cooled approach to a liquid backplane solution can be associated with large facility-level changes under specific assumptions (e.g., PUE 1.80 → 1.23 at 2,262 kW IT). Those figures should be treated as scenario-based facility results, not as an RDHx-only capacity claim.
Practical next step
If you want to reduce procurement ambiguity, the most effective next step is a one-page RDHx sizing and commissioning worksheet that captures:
sustained / 95th / peak rack kW
water supply temperature band and dew point margin
flow budget per door
acceptance criteria for heat-to-water performance and alarms
If you’re planning in the 20–40 kW band, Coolnetpower’s engineering framing of RDHx as an engineered interface (not a bolt-on accessory) is a useful companion read: power and cooling per rack best practices for 20–40 kW (2026).
