For those managing 1u Servers, balancing thermal performance with noise levels is a common challenge. This exploration delves into practical methods for optimizing fan speeds in a 1U server environment, aiming to reduce noise and enhance power efficiency without compromising system stability. The findings are based on experimental trials using a fan control script and meticulous monitoring of server performance under varying loads.
Utilizing a script (https://github.com/petersulyok/smfc) to govern fan speeds proved effective in significantly lowering server noise. The server, now operating within a home environment, achieves a much more tolerable noise level, hovering around 50 dB or less for much of the time. Fan speeds were adjusted to approximately 4500 RPM for Delta fans and 1800 RPM for Dynatron fans, representing a significant reduction from their maximum capabilities.
Through careful observation of IPMI fan speed readings, the operational ranges for both Delta and Dynatron fans were determined. Delta fans can range from 2400 RPM to a high of 15000 RPM, while Dynatron fans operate between 1300 RPM and 8000 RPM. The fan control script was configured with these thresholds to maintain optimal cooling. Consequently, CPU temperatures at idle remain consistently between 32 to 35 degrees Celsius and do not exceed 60 degrees Celsius even under full stress tests. The 10 GbE NIC temperature typically stays around 53 degrees Celsius, dropping to 49 degrees Celsius during idle periods.
Further testing was conducted under different load conditions to assess the impact of fan speed optimization on power consumption. The server configuration for these trials included:
- Dual E5-2680v4 CPUs
- 128GB RAM (varied in some tests)
- 1 x 10 GbE NIC
- Three SSDs (no dedicated SSD fans)
- No additional PCIe cards beyond the NIC
- Redundant 2 x 750W power supplies
Fan speed modes tested included “Optimal” (4500 RPM Delta, 1800 RPM Dynatron) and “Full Speed”. Power consumption was measured using both a wattmeter and IPMI readings.
| No. of CPUs | RAM | No. of Fans (Delta + Dynatron) | 10 Gb Quad NIC | Power (Idle/ Low Load/ Medium Load) in watt |
|---|---|---|---|---|
| 2 | 128GB | 6 + 2 → Optimal mode | Yes | 105/ 140/ 220 |
| 2 | 128GB | 6 + 2 → Full speed mode | Yes | 200/ 240/ 380 |
| 2 | 128GB | 6 + 2 → Heavy IO mode | Yes | 180/ 220/ 360 |
| 2 | 128GB | 6 + 2 → Optimal mode | No | 85/ -/ – |
| 1 | 64GB | 6 + 1 → Optimal mode | No | 61/ -/ – |
| 1 | 64GB | 3 + 1 → Optimal mode | No | 58/ -/ – |
| 1 | 64GB | 3 + 1 → Optimal mode | Yes | 74/ 100/ 180 |
The data clearly demonstrates the power savings achieved through optimal fan speed control compared to full speed operation. Even in “Heavy IO mode,” power consumption remains significantly lower than full speed, while maintaining acceptable temperatures.
It’s also noted that the 1U server consumes approximately 20 watts when powered off but with IPMI accessible. A continuous beep is observed if a single power supply is disconnected, even with a CPU removed, indicating the necessity of both PSUs for proper operation as per Supermicro’s documentation and precluding redundant PSU functionality in this specific configuration. These findings underscore the importance of balanced fan control for noise reduction and power efficiency in 1U servers, crucial for deployments in noise-sensitive environments.
