Rear Door Heat Exchanger Capacity Is Capped by Rack Airflow

An Open Compute Project study credits 3,500 cfm of rack airflow, heated 15°C by the servers, with 29.5 kW of IT load. Run through Q = ρ × cp × V × ΔT with air at 1.16 kg/m³ and 1,007 J/(kg·K), that airflow (1.652 m³/s) carries 28.9 kW, within 2% of the published figure. A rear door heat exchanger (RDHx) is a water coil that replaces a rack’s back door and pulls heat out of the server exhaust, so it can only remove what that air stream delivers. How far an RDHx reaches in density is set first by the air crossing it and the water feeding it, and published ratings state those two inputs in very different ways.

Seven door capacity figures rest on seven sets of conditions

Two build types appear in the documents. A passive door has no fans and relies on the server fans to push exhaust through the coil. An active door adds its own fans to overcome the coil’s resistance. Both are sold with a single kW number on the front page.

Vertiv’s guide specification for the passive Liebert DCD rates the DCD35 at 35 kW and the DCD50 at 50 kW of net sensible cooling.

Both ratings assume 21°C ambient air, 12°C entering water and 50% relative humidity.

Source: Vertiv, “Vertiv Liebert DCD 25-50kW Water-Cooled Rear Door Heat Exchanger Guide Specification” (June 2022)

Lenovo’s documentation for its passive Rear Door Heat eXchanger V2 gives no single rating. It publishes performance curves for 30 kW and 20 kW loads at 27°C rack inlet air, with the share of heat removed depending on the water temperature and flow the operator selects. The same page ties water temperature to the room, asking for supply above dew point unless the water system measures dew point and adjusts to it.

Source: Lenovo Docs, “Rear Door Heat eXchanger V2”

Supermicro’s datasheet lists active door models whose ratings run from up to 20 kW to up to 120 kW. That top model carries six fans and a 3.6 kW maximum power draw. No entering water temperature, air temperature or airflow is printed for any model on the sheet.

The Open Compute Project’s door requirements document proposes one reference point for rating capacity: 16°C water entering, air arriving at the door at 45°C and leaving at 30°C, at sea level.

Source: Open Compute Project, “ACS Door Heat Exchanger Requirements for Open Rack, Revision 1.0” (June 2021)

A figure with no stated water and air temperatures cannot be placed on the same axis as one that has them. The 120 kW number describes the top of a product range, not a rating that can be set beside a 50 kW figure measured against 12°C water.

Table 1. One capacity figure per door, under the conditions each source stateskW per door

* “Up to” value and curve load, not ratings at a common condition

** DCD50 technical table lists 23°C server supply air

*** Per-door average: 31.6 kW over six doors; rack load 10 to 11 kW

Source: Super Micro Computer, “Rear Door Heat Exchanger (RDHx) Datasheet” (2026); Vertiv, “Liebert DCD Guide Specification SL-70960” (June 2022); Lenovo Docs, “Rear Door Heat eXchanger V2”; Open Compute Project, “Evaluating the Limits of Door Heat Exchanger Solutions Optimized for Open Rack V3”; LBNL for U.S. DOE FEMP, “Data Center Rack Cooling with Rear-door Heat Exchanger” (June 2010)

Rack airflow runs out ahead of the coil

Every door in Table 1 cools air, so its ceiling is the heat that air can carry. Rearranging the lead equation gives the airflow a rack needs, V = Q ÷ (ρ × cp × ΔT). With the same air properties and a 15 K rise across the servers, the results are upper bounds on what the door is asked to see:

• 30 kW needs 1.71 m³/s, or 3,630 cfm
• 50 kW needs 2.85 m³/s, or 6,050 cfm
• 132 kW needs 7.53 m³/s, or 15,960 cfm

The 15 K rise is the OCP study’s own assumption. Its facility tables give a server air rise range of 15°C to 23°C at 35°C inlet, so a hotter exhaust buys only a limited amount of extra capacity per cfm.

The 132 kW row is the rack class described in the GB200 rack density figures. Carrying that load as air through a door would take more than four times the 3,500 cfm the OCP study worked with, through the same rack face and the same server fans.

Active doors add airflow, but the numbers stay tight. Vertiv rates the fan module for the DCD50 at 9,000 m³/h with N+1 fan redundancy, which converts to 2.5 m³/s. Solving the same equation at that airflow:

• Air rise needed for 50 kW = 50,000 ÷ (1.16 × 1,007 × 2.5) = 17.1 K
• Heat carried at a 15 K rise = 1.16 × 1,007 × 2.5 × 15 = 43.8 kW

The coil may be good for 50 kW. Whether it ever sees 50 kW depends on the servers delivering a hotter exhaust or on the fans running past their redundant rating, and neither condition appears on the rating line.

Fans are also the running cost of that air. The OCP requirements document asks that an active door draw no more than 2% of its rated capacity, excluding fan-failure operation.

Vertiv lists a 1,185 W maximum input for the DCD50 fan module, which is 2.4% of 50 kW.

Supermicro lists a 3.6 kW maximum for its 120 kW model, or 3.0%. Both are maximum draws rather than draws at rated capacity, so neither is a direct test against the OCP guideline.

Source: Vertiv, “Liebert DCD Guide Specification SL-70960” (June 2022); Super Micro Computer, “Rear Door Heat Exchanger (RDHx) Datasheet” (2026)

Warmer water turns a 58% coil job into a 79% one

The OCP study fixes the load at 29.5 kW and rack airflow at 3,500 cfm, then asks what supply water each door accepts while returning air to the room at 30°C.

With a 10 K water rise, the off-the-shelf door needs 19°C water. The door sized to Open Rack V3, with a larger face and more tube rows, runs at 26°C.

Source: Open Compute Project, “Evaluating the Limits of Door Heat Exchanger Solutions Optimized for Open Rack V3”

The reason is the initial temperature difference (ITD), the gap between air entering the coil and water entering it. The door must cool 45°C exhaust back to 30°C whatever the water temperature. In the study’s cases air is the stream with the lower heat capacity rate, since it changes 15 K against the water’s 10 K for the same heat, so the fraction of the ITD the coil must close is its air-side effectiveness, ε = (air in − air out) ÷ ITD:

• OCP reference condition, 16°C water: 15 ÷ (45 − 16) = 0.52
• Off-the-shelf door, 19°C water: 15 ÷ (45 − 19) = 0.58
• Door sized to Open Rack V3, 26°C water: 15 ÷ (45 − 26) = 0.79
• Same door at 27°C water: 15 ÷ (45 − 27) = 0.83

These are the effectiveness values each case requires, derived from the study’s stated temperatures. They are not measured coil values.

Seven degrees of warmer water pushed the required effectiveness from 0.58 to 0.79, and the study reached it with a bigger, deeper coil. A door bought off the shelf and a door sized to its rack are different products even when both are listed at the same kW, and the difference shows up as chiller hours.

Colder water is not a free way back. The OCP requirements document recommends not dropping below 16°C to avoid condensation in the door, and Vertiv’s DCD specification caps room dew point at 10.5°C. Lenovo asks for water above dew point unless the supply tracks it.

An RDHx therefore works inside a window. The room dew point sets the bottom, and the amount of coil that fits in a rack door sets how warm the water can run at the top.

Why does a rear door end up as the residual-heat stage?

The best-documented retrofit cooling case is a Lawrence Berkeley National Laboratory demonstration published through the DOE Federal Energy Management Program. Passive doors were fitted to six racks drawing 10 to 11 kW each.

Water came from a cooling tower loop through a plate-and-frame heat exchanger, nearly eliminating chiller energy for those racks. It entered the doors at 22°C and left at 24.4°C, at 34 L/min per door.

The bulletin’s arithmetic is in US units; the per-door average is added here:

• Heat removed: 500 × 54 GPM × (76°F − 72°F) = 108,000 Btu/h = 31.6 kW
• Share of server load: 31.6 kW ÷ 66 kW = 48%
• Average per door: 31.6 kW ÷ 6 = 5.3 kW

Source: Lawrence Berkeley National Laboratory for the U.S. DOE Federal Energy Management Program, “Data Center Rack Cooling with Rear-door Heat Exchanger” (June 2010)

Air left the doors near 26.7°C against a 21°C server inlet, and the existing CRAC units carried the rest. The low capture follows from choosing tower water over chilled water, not from a weak coil, and it bought chiller-free removal of nearly half the heat.

The bulletin also reports doors that did not seal tightly, hot air short-circuiting inside racks, and poor fit on racks more than 10 years old. Those fit-and-seal problems are the same ones that decide most attempts to add liquid cooling without rebuilding the hall.

The OCP study’s route to higher density keeps the door and moves most of the heat off it. Its hybrid case pairs the 30 kW door with an in-rack coolant distribution unit of 50 kW feeding cold plates, for 80 kW per rack.

Cold plates here mean direct liquid cooling (DLC), in which coolant takes heat from processors through plates bolted to them, following the cold plate path to facility water. The study runs both the door and the DLC loop on 27°C facility water. Its CDU capacity is the authors’ feasibility estimate, not a tested unit.

Supermicro’s datasheet describes its door range in the same role, as removing the residual heat left in the rack. The OCP study, for its part, describes the air-assisted door as a way to ease the move toward liquid-cooled facilities.

That is the sense in which the door is a bridge. As the primary cooling for an air-cooled rack, it stops where rack airflow stops, and the doors rated at stated conditions sit between 29.5 kW and 50 kW. As the stage that catches whatever DLC leaves in the air, it keeps a job at any density.

A door rating is one point on a surface of water and air temperature

Each capacity figure in Table 1 is a single point on a surface defined by water temperature, air temperature and airflow. Two doors listed at the same kW can sit far apart on that surface, as the 7 K gap in the OCP study shows.

The calculations here hold air properties constant at sea level and treat the coil as one effectiveness value with uniform airflow, while real doors see uneven exhaust, bypass around the frame and fan curves that shift as the coil loads. The OCP comparison is a design study built on performance curves, the LBNL case is a 2010 installation on servers of that era, and the vendor ratings are manufacturer figures without independent test reports attached. Any door selection or piping change needs review by a qualified mechanical engineer against the actual room, servers and water plant.

Five items put two door sheets on the same footing:

• Entering water temperature and water rise behind the kW figure
• Air temperature arriving at and leaving the door
• Airflow the rating assumes, and whose fans supply it
• Room dew point against the lowest planned water temperature
• Fan power at rated capacity, not maximum draw

References

• Evaluating the Limits of Door Heat Exchanger Solutions Optimized for Open Rack V3 / Open Compute Project / date not printed on the document / https://www.opencompute.org/documents/acs-door-hx-whitepaper-final-230419-pdf
• ACS Door Heat Exchanger Requirements for Open Rack, Revision 1.0 / Open Compute Project / June 15, 2021 / https://www.opencompute.org/documents/acs-door-hx-open-compute-requirements-for-open-rack-rev1-0-1-pdf
• Data Center Rack Cooling with Rear-door Heat Exchanger, Technology Case-Study Bulletin / Lawrence Berkeley National Laboratory for the U.S. DOE Federal Energy Management Program / June 2010 / https://datacenters.lbl.gov/sites/default/files/rdhx-doe-femp.pdf
• Vertiv Liebert DCD 25-50kW Water-Cooled Rear Door Heat Exchanger Guide Specification, SL-70960 Rev A / Vertiv / June 2022 / https://www.vertiv.com/494452/globalassets/products/thermal-management/high-density-solutions/vertiv-liebert-dcd-25-50kw-water-cooled-rear-door-heat-exchanger-guide-specification-sl-70960.pdf
• Rear Door Heat Exchanger (RDHx) Datasheet / Super Micro Computer, Inc. / 2026 / https://www.supermicro.com/datasheet/datasheet_Rear_Door_Heat_Exchanger.pdf
• Rear Door Heat eXchanger V2, ThinkSystem Heavy Duty Rack Cabinets / Lenovo Docs / date not stated / https://pubs.lenovo.com/hdc_rackcabinet/options_42u_heat_exchange_door