Why RDHx Fan Selection Is More Complex Than Server Fan Selection
Server fans push air through a known chassis resistance — the system curve is relatively predictable. RDHx fans face a two-part challenge: they must overcome both the inherent resistance of the water-cooled door panel AND compensate for variable server exhaust conditions. Getting this wrong means either insufficient cooling (thermal throttling on GPU workloads) or oversized fans that waste energy and generate excessive noise in the data hall.
This guide provides a systematic approach to RDHx fan selection for AI data center deployments, from measuring system resistance to specifying N+1 redundant arrays.
Understanding the RDHx System Resistance Curve
The RDHx door panel creates airflow resistance through its heat exchanger fins. This resistance follows a square-law relationship with airflow — double the airflow, quadruple the pressure drop. Typical values:
- Low-density fins (standard cooling): 80–150 Pa at rated flow
- Medium-density fins (AI server density): 150–250 Pa at rated flow
- High-density fins (maximum heat rejection): 250–350 Pa at rated flow
To find the operating point, plot your fan’s P-Q curve (pressure vs. airflow) and your system resistance curve on the same graph. The intersection is the operating point. For reliable operation across varying server loads, the fan operating point should remain above the system resistance curve with ≥20% margin even at minimum PWM speed.
Fan Array Design for 42U AI GPU Racks
A 42U rack running 8× NVIDIA H100 nodes can dissipate 50–60 kW. RDHx panel sizing must match this thermal load at the available chilled water temperature (typically 14–18°C supply). Fan array design:
| Fan Size | CFM/Fan | Fans Required (N) | N+1 Config | Total CFM |
|---|---|---|---|---|
| 92×25mm | 80 | 14 | 16 | 1,280 |
| 120×25mm | 110 | 10 | 12 | 1,320 |
| 120×38mm | 140 | 8 | 10 | 1,400 |
N+1 redundancy means one fan can fail without the array dropping below minimum required airflow. For AI workloads with continuous high thermal output, N+2 (two fan failures tolerated) is increasingly specified.
Static Pressure Requirements by Application
Not all RDHx installations have the same static pressure requirement. Key variables include chilled water temperature, fin density, door panel depth, and whether an air curtain or plenum is used:
- Standard enterprise RDHx: 200–280 Pa required — 92mm fans (350 Pa rated) provide adequate margin
- High-density AI GPU racks (30–60 kW): 280–350 Pa required — 120mm fans (380–420 Pa) recommended for comfortable margin
- Ultra-high-density (>60 kW/rack): 350–420 Pa required — 120×38mm fans or dual-layer fan arrays
PWM Control and Fan Speed Optimization
RDHx fan arrays should use PWM control with tachometer feedback to dynamically match fan speed to actual server heat output. Best practice control architecture:
- Monitor server inlet temperature (target: ≤25°C)
- Monitor RDHx supply water temperature
- Adjust fan PWM duty cycle to maintain target inlet temperature
- Set minimum PWM at 20–30% (never below stall speed) for bearing lubrication
- Implement alarm on tachometer signal loss (fan failure detection)
Fengheng Technology RDHx Fan Specifications
| Model | Size | Speed | Airflow | Static Pressure | Voltage | MTBF |
|---|---|---|---|---|---|---|
| FH-9225-48V | 92×25mm | 8,000 RPM | 80 CFM | 350 Pa | 48V DC | ≥50,000 h |
| FH-1225-48V | 120×25mm | 5,500 RPM | 110 CFM | 380 Pa | 48V DC | ≥50,000 h |
| FH-1238-48V | 120×38mm | 4,500 RPM | 140 CFM | 420 Pa | 48V DC | ≥50,000 h |
All models: 4-pin PWM, tachometer output, dual ball bearing, CE/RoHS certified, operating temperature -10°C to +70°C.
RDHx Fan Samples Available
Fengheng Technology offers 92mm and 120mm RDHx fans with 48V DC input for AI data center qualification. Sample lead time: 2 weeks. Contact [email protected] or view the full liquid cooling fan portfolio →