Supermicro has expanded its liquid cooling lineup with ten heat exchangers for rack rear doors, known as RDHx. The new models offer capacities ranging from 10 to 120 kW per unit and are designed to adapt existing data centers for artificial intelligence and high-performance computing servers without a complete overhaul of their technical spaces.
The key points of Supermicro’s RDHx cooling in 30 seconds
- The range includes ten models with cooling capacities from 10 to 120 kW.
- Supermicro announces configurations capable of managing up to 240 kW per rack.
- The doors can be installed on EIA racks, Open Rack v3, and Nvidia MGX systems.
- They can also be combined with direct-to-chip cooling for higher-density loads.
- The proposal aims to modernize existing data centers with less construction work than other liquid cooling solutions.
The new heat exchangers are part of Data Center Building Block Solutions (DCBBS), Supermicro’s offering to supply servers, racks, networking, power, cooling, management software, and deployment services as a pre-integrated infrastructure.
This expansion arrives as new generations of GPUs generate more electrical power and heat per rack. Traditional air cooling may still be sufficient for conventional servers, but it becomes less effective as density increases, requiring larger volumes of air moved by fans and HVAC systems.
An RDHx seeks to address part of this problem without delivering liquid directly to each processor. It replaces the rear door of the rack with a heat exchanger that captures the hot air expelled by servers, transfers that heat to a liquid circuit, and returns cooler air to the room.
A cooled door to upgrade existing data centers
The main advantage of rear-door heat exchangers is that they can be added to air-cooled servers. Operators don’t need to replace all the machines with cold-plate systems or run piping inside each chassis.
Supermicro presents this option as a way to increase computing density with fewer infrastructure changes. The installation still requires a circuit to transport heat out of the rack, hydraulic connections, control systems, and enough capacity to cool the fluid. The difference is that the intervention is concentrated in the cabinet and its rear door.
| Technology | How it removes heat | Changes in servers | Most common use |
|---|---|---|---|
| Air cooling | Fans and room HVAC | None | Low to medium density racks |
| RDHx | Cools hot air exiting the rack | Usually none | Modernizing existing rooms |
| Direct-to-chip | Delivers liquid to CPU and GPU plates | Servers prepared for liquid | High-density AI and HPC |
| Immersion | Submerges equipment in dielectric fluid | Specific hardware and tanks | Very dense loads or specialized setups |
Supermicro’s technical documentation describes the RDHx as the least disruptive alternative among its liquid cooling options, emphasizing that it can be installed later. It also reminds that the hot liquid must be cooled before returning to the door, so external thermal infrastructure remains necessary.
The ten new models range from 10 to 120 kW per door. The company also mentions configurations up to 240 kW per rack, though the announcement does not specify how many exchangers or which cooling combination is needed to reach that maximum.
| Features of the new RDHx range | Announced data |
|---|---|
| Number of models | 10 |
| Minimum capacity | 10 kW |
| Maximum capacity per unit | 120 kW |
| Maximum capacity per rack | Up to 240 kW |
| Compatible racks | EIA, ORv3, MGX |
| Available power | AC or DC |
| Management | Redfish, SNMP, web interface, SuperCloud Composer |
| Redundancy | N+1 |
| Additional protection | Anti-condensation control |
Lower-capacity models can be used in enterprise racks or specific zones that are beginning to exceed existing cooling capabilities. Higher-end versions target accelerated servers, AI clusters, and HPC systems.
A thermal capacity of 120 kW does not mean all racks must consume that power. It indicates how much heat the system can remove under certain conditions. Actual performance will depend on fluid temperature and flow rate, ambient air temperature, fan speed, and data center configuration.
An intermediate solution between air and direct-to-chip
The RDHx occupies an intermediate position. It keeps servers cooled by air but uses liquid to transfer heat after it leaves the machines. This can lessen the load on room cooling systems and limit hotair recirculation.
Its capacity may be insufficient on its own when CPU, GPU, memory, and networks concentrate several hundred kilowatts. That’s why Supermicro will allow combining doors with its direct-to-chip solutions, also called D2C.
In a D2C system, coolant circulates through plates directly attached to components like processors and accelerators. The liquid absorbs heat near its source and transports it to a refrigerant distribution unit or CDU.
The rear door can then collect residual heat generated by sources not connected to the liquid circuit, such as power supplies, storage units, networking components, or regulators.
| Configuration | Main heat removed by liquid | Remaining heat |
|---|---|---|
| Just RDHx | Hot air expelled by servers | Depends on the door’s capacity |
| Only direct-to-chip | CPU, GPU, and other components with cold plates | Must be evacuated by air or other systems |
| D2C plus RDHx | Main components and residual air | Lower thermal load on the room |
| Full D2C cooling | Compatible with CPU, GPU, memory, network, and power supplies | Can capture most of the heat |
Supermicro states that its second-generation direct-to-chip solutions can capture up to 98% of the heat when including CPU, GPU, memory, PCIe switches, regulators, and power supplies. The company also reports potential reductions of up to 40% in energy and water use, and up to 20% in total cost of ownership. These estimates are based on the full DLC-2 architecture and should not be directly applied to systems using only RDHx doors.
Currently, the company offers coolant distribution units up to 250 kW inside the rack and up to 1.8 MW in in-row configurations. These CDU circulate fluid, control its temperature, and connect the equipment’s circuit to the building’s thermal infrastructure.
Management, redundancy, and rack compatibility
Supermicro intends to sell these heat exchangers as part of a complete platform rather than as isolated accessories. Customers will be able to purchase the door along with accelerated servers, power distribution, networking, rack integration, and installation services.
The models will be compatible with conventional EIA racks, Open Rack v3 (ORv3), and Nvidia MGX systems. This variety is important because data centers do not use a single geometry: traditional enterprise racks coexist with open designs and platforms optimized for accelerators.
DC power versions can connect to rack busbars, while AC versions facilitate installation in rooms with standard power supply.
The range features intelligent fan control, N+1 redundancy, and condensation protection. The latter is necessary because overly cold liquid can cause moisture in the air to condense on surfaces and connections.
| Control functions | Purpose |
|---|---|
| Adjustable fans | Modulate airflow according to thermal load |
| N+1 redundancy | Maintain cooling if a fan or component fails |
| Anti-condensation protection | Prevent moisture around equipment and ducts |
| Temperature sensors | Monitor air and liquid circuit temperature |
| Pressure and flow measurement | Detect restrictions, leaks, or pump failures |
| Pump supervision | Ensure proper circuit operation |
| Remote management | Integrate alerts and telemetry into data center operations |
Operators will be able to monitor temperature, pressure, flow, and pump status via Redfish, SNMP, a web interface, and SuperCloud Composer.
Supermicro’s management platform already allows for monitoring thermal profiles of GPUs, CPUs, memory, distribution units, power supplies, and cooling towers. Adding RDHx expands this visibility to another part of the thermal circuit.
Monitoring is crucial. A 100 kW rack can generate an amount of heat in one hour equivalent to the energy consumed by the equipment itself. A flow restriction, a stuck fan, or a misconfigured valve can quickly lead to temperature rises and automatic performance reductions.
Why RDHx is interesting for old data centers
New AI deployments can be designed from the start with piping, CDU, cooling towers, and servers prepared for liquid. The problem appears in existing data centers, built for racks that consumed far less power.
Upgrading an entire room can require raised floors, new piping, replacing chillers, and service interruptions. Rear-door solutions allow concentrating adaptation in specific rows or racks, enabling a progressive migration.
This does not turn any room into a 240 kW-per-rack data center. Electrical infrastructure, liquid distribution, weight, floor load, and heat rejection capacity must support the new load as well.
| Element to review | Reason |
|---|---|
| Electrical capacity | An AI rack can require tens or hundreds of kilowatts |
| Cooling circuit | Must carry and dissipate all captured heat |
| Floor and structure | GPU systems and liquid components increase weight |
| Electrical wiring | Requires suitable busbars, PDU, and protections |
| Leak detection | Limits risk near electronics |
| Rear clearance | The door and connections need space to open |
| Maintenance | Fans and components should be replaceable without shutting down the rack |
This announcement broadens Supermicro’s options for operators needing more powerful GPUs while avoiding complete infrastructure replacement for direct-to-chip cooling. Effectiveness depends on the specific room conditions and how much heat can be actually removed with available water or refrigerant.
The figure of up to 240 kW per rack reflects market direction: cooling can no longer be treated as a separate auxiliary system. In new AI factories, servers, network, power, and thermal control must be designed as a unified system.
Frequently Asked Questions
What is a rear-door heat exchanger?
It’s a door installed behind the rack that cools the hot air expelled by servers. The heat passes to a liquid circuit before the air returns to the room.
Do RDHx units contain liquid inside the servers?
Not necessarily. They can be used with air-cooled servers, although Supermicro also allows combining them with direct-to-chip systems.
Can an RDHx be installed in an existing data center?
Yes, this is one of its main advantages. However, installing it requires hydraulic connections, heat dissipation capacity, and enough space around the rack.
What does a 120 kW capacity mean?
It’s the maximum amount of heat the exchanger can remove under specific conditions. It does not directly indicate actual rack power consumption nor guarantee that performance in any installation.
