How to Choose an External Integrated Radiator for AI Server Cabinets
Introduction: Four numbers decide whether an external integrated radiator belongs on your AI server cabinet: heat load, pump flow and head, port standard, and footprint.
When a multi-GPU cabinet runs out of room inside the chassis, the problem stops being about picking a better fan. The heat is already there, and the rack simply cannot move it out fast enough through sealed air paths. Thermal engineers hit this wall when accelerator cards, power stages, and network cards stack up in a closed enclosure and internal airflow collapses. Moving heat rejection outside the cabinet changes the geometry of the problem. The four inputs below are the ones that decide whether a specific external unit fits your build.
Start With Cabinet Heat Load and External Radiator Capacity
Sizing starts with a wattage total, not a product photo. Add the sustained heat output of every device you plan to put on the loop: GPU cards, CPUs, VRM and power stages, network cards, and any accelerator or FPGA boards. Use sustained figures from vendor datasheets rather than idle numbers. AI workloads swing hard, so design around the peak training or inference burst instead of the daily average. Intel's processor support documentation is a practical reference for how much cooling a high-TDP part expects before you scale that expectation up to a full card population. ASHRAE's data center resources make the same case at rack level, which is why liquid cooling moved from optional to standard in high-density compute rooms. The OCOCOO BC5-kit is built around a nominal 4000W design heat load. Treat that as your sizing budget: if the devices on your loop add up below it, you have headroom for peak bursts and for warmer room air. Room temperature matters as much as wattage, because the temperature difference that drives heat into the air shrinks as intake air gets hotter, and a coil that rejects 4000W in a 25°C room rejects less at 35°C. The radiator, eight PWM fans, and pump in the BC5-kit are matched as one unit, which is where an external integrated water cooling radiator differs from assembling a bare coil, a fan wall, and a separate pump yourself.
Match Pump Flow and Head to the AI Server Loop Resistance
Flow and head are the two numbers that decide whether coolant actually reaches the far end of a multi-block loop. Flow measures volume moved per hour; head measures the resistance the pump can overcome, expressed in meters of liquid column. They move in opposite directions: as you add cold plates, quick-disconnects, and tubing bends, resistance rises and delivered flow drops. The Engineering Toolbox reference on pump head and pressure explains that relationship, and it is the reason a pump rated at high flow on a bench can look modest once it is pushing through six GPU blocks in series.
1. What 1300 L/H Flow and 5M Head Mean for Multi-Block AI Server Loops
The SC-P90/P90D pump inside the BC5-kit carries a 1300 L/H maximum flow rating and 5M maximum head at 4500 RPM. Those are peak figures at low resistance, which is how every pump is specified. In a real AI server loop with several GPU cold plates, a CPU block, quick-disconnects, and a few meters of tubing, delivered flow will sit lower than the ceiling, and 5M of head is what keeps that drop manageable. As a practical rule, keep the loop order simple: reservoir to pump, pump to cold plates, cold plates to the radiator, radiator back to the reservoir. Fewer fittings and shorter runs mean more of that head budget goes into moving coolant instead of fighting its own path.
2. How MCU-Controlled Three-Phase Brushless Pump Speed Supports Variable AI Loads
The pump runs on a three-phase brushless DC motor with MCU control, powered from DC 12V/1.8A. Electronic commutation keeps brushes out of the wear path and lets the controller hold speed steady as loop resistance and fluid temperature shift during a long training run. That suits AI hardware, where loads are never flat: inference idles near the floor, then training spikes for hours. An MCU-controlled pump paired with PWM fans lets you match cooling effort to real demand, quiet and efficient under light load and at full flow when the accelerators ramp. Pump noise is rated at ≤25 dB, and the eight fans remain the dominant sound source, so overall noise tracks fan speed.
Match G1/4 Ports, Footprint, and Fan Control to Cabinet Space
G1/4 threaded ports are the working standard in workstation and server liquid cooling, so the BC5-kit accepts widely available fittings, tubing, and cold plates from multiple suppliers. Fittings, tubing, coolant, and water blocks are separate items, so confirm thread depth, tubing inner diameter, and block compatibility for your specific loop before ordering. The radiator is aluminum, which sets a materials rule for the whole circuit: use an inhibited coolant rated for aluminum, and give mixed-metal loops proper protection when copper blocks share the same reservoir. That is standard practice for any industrial water cooling solution that pairs an aluminum coil with copper cold plates. Footprint and mounting decide whether the unit fits the room, not just the loop. At 592 × 243 × 375 mm and roughly 8.20 kg with packaging, the BC5-kit needs a stable surface beside the rack or on a shelf, with unrestricted air on the fan side and enough clearance for service. The transparent reservoir makes level checks quick, and the manual pressure relief valve handles thermal expansion inside a closed water cooling system solution where pressure builds with temperature. Control is handled two ways: a speed knob on the unit and PWM for the fans. If a standard 592 mm length does not match your cabinet side, OCOCOO supports DfM work on radiator length, fin density, and logo as an external radiator manufacturer.
Conclusion
The four inputs are easy to hold in your head: total sustained watts with peak headroom, pump flow and head measured against your loop resistance, G1/4 port compatibility for fittings and blocks, and a footprint that fits your floor or cabinet side. Once those numbers are on paper, choosing among external liquid cooling solutions stops being guesswork. The OCOCOO BC5-kit covers a nominal 4000W design heat load with eight PWM fans, a 1300 L/H pump, and a 592 × 243 × 375 mm chassis. Send your heat load, port plan, available footprint, and target flow to the factory team for a quote, catalog, and confirmation of fittings, tubing, and coolant.
FAQ
Q：How do I know if an AI server cabinet needs an external integrated radiator instead of an internal AIO?
A：Compare available internal volume against total wattage. If your cabinet holds multiple accelerators and the inside is already crowded with cards, power supplies, and cable runs, an internal AIO has nowhere to reject heat except back into the same enclosure, so intake air keeps warming up. An external integrated radiator moves the coil, fans, and reservoir outside the rack and drops cabinet-side air temperature back toward room level. If your sustained load climbs into the multi-kilowatt range and internal airflow cannot clear it, external rejection is the practical answer.
Q：What do 1300 L/H flow and 5M pump head mean for a multi-block AI server cooling loop?
A：They describe the pump's ceiling, not what every loop receives. The 1300 L/H maximum flow is measured at low resistance; every cold plate, quick-disconnect, and tubing bend you add increases resistance and trims delivered flow. The 5M maximum head is the pressure budget that lets the pump push through several blocks in series and still return coolant to the reservoir. Keep the loop short, use full-bore fittings, and avoid unnecessary adapters so more of that head goes into circulation.
Q：Can the OCOCOO BC5 external integrated radiator connect to standard G1/4 fittings and fit a cabinet-side footprint?
A：Yes. The BC5-kit uses G1/4 threaded ports, the common standard for workstation and server liquid cooling, so standard fittings, tubing, and cold plates connect directly. The unit measures 592 × 243 × 375 mm and weighs about 8.20 kg with packaging, which suits a shelf or floor position next to the cabinet with clear airflow on the fan side. Confirm your tubing size, block thread, and coolant choice before integration, and ask about radiator length or fin density options if your cabinet side has a tight envelope.
Sources / References
ASHRAE Data Center Resources | Data Centers
Pump Head and Pressure: Conversion, Calculations, and Charts
Comments
Post a Comment