Building a Silent Workstation

The purpose of this paper is to describe how to build a silent workstation. By silent, I mean inaudible in an office with a closed door, or in a suburban home. These are locations with fairly low background noise levels (40dBA - 45dBA). Lowering the noise level of a computer, especially a high-performance computer, to be inaudible under these conditions is a significant task. Keeping the component temperatures low enough to deliver a long life can be even more difficult.

1. Components
2. Assembly
3. Temperatures
4. Noise
5. Performance
6. Mounting the SSD
7. Alternate Motherboards
8. Alternate CPUs
9. Painting it Green
10. Issues Encountered

Last updated - September 3, 2009

NOTE: This web page is a Work-In-Progress. It is NOT complete.

Temperatures

The motherboard was installed in the case, along with the video card and the disks. It looks fine, and runs well, so we ran a temperature check. The video card and CPU were fine, but the X58 chip (to the right of the CPU) had a heatsink temperature of 76C. Considering that this is an idle temperature, that's too hot for good long-term reliability, so I need to replace the heatsink.

The low temperatures of the video card and CPU are due to the improvised ducting that was used in this prototype system. A small piece of cardboard can be seen going between the CPU heatsink and the power supply. This forces the majority of the air to go through the CPU heatsink, or the video card heatsink. A small piece of foam, between the video card heatsink and the case, closes the last significant airflow path. The side cover of the case is close enough to the CPU heatsink and the cardboard to not create a significant airflow issue.

The chipset heatsink has now been replaced with a Thermalright HR-05/IFX. It was a tight fit, but the chipset temps are now under control.

A close-up showing just how tight these heatsinks are.

The disk temperatures were a bit too warm, so a pair of Arctic Cooling AF 9225 PWM fans were added in the front of the case. The disk temperatures dropped to normal, but the motherboard didn't show any speed for either fan. The fans were checked, and found to not have a functional tachometer signal. The good people at FrozenCPU rushed a pair of Scythe Kama PWM 92MM fans out to replace the broken fans, and all is well, but a bit louder.

If you can get some functional Arctic Cooling fans, the system will be quieter. The Scythe Kama fans are a bit louder, and move more air. This is a tradeoff. The difference in noise level is only 2-3dBA, but it's enough to push the system above the "silent" status for many people.

The inlet, exhaust and power supply exhaust temperatures were measured with a Fluke 61 infrared thermometer. The system temperatures and fan speed were measured by the Supermicro system monitoring utility Super Doctor III. The CPU and GPU temperatures were measured by RealTemp. The power from the wall was measured with a Kill-A-Watt.

The CPU load was Prime95. The GPU load was FurMark.

It can be seen that the fan speed has no impact on the power usage, and only a slight impact on the temperatures. It does, however, have an impact on the noise. Unless the small difference in airlow makes the difference between acceptable temperatures and unacceptable temperatures, it is suggested that the fans be set for Workstation.