The INQ guide to silencing a PC

WHAT DOES IT take to make a modern PC quiet? To find out, I attempted to build a very quiet, high-performance PC using components from various manufacturers. I set myself a challenge by basing the system on an extremely hot 3.4 Ghz Pentium 4, and by trying to keep things as simple as possible - so no watercooling.

The significant sources of noise in PCs are, in approximate order of loudness, the CPU fan, the graphics card fan, the CD or DVD drive, the power supply fan and case fan, and the hard disk drive.

Most of the noise, then, comes from fans. There are four ways to make fans quieter: reduce the speed, use a larger fan, minimize obstructions which block the airflow, or remove the fan altogether. If you're not familiar with the basics of PC noise control, I'd recommend silentpcreview.com as a good source of useful information like this.

Reducing the fan speed is simple, and if the speed is automatically varied based on temperatures inside the PC, we can keep the noise down, but still call on additional cooling power when we need it. Commonly-used mainboard I/O chips have had the ability to automatically control fan speeds for several years, but until recently most mainboard manufacturers didn't put in the necessary connections and driver electronics to translate the control signals from the I/O chips into voltages that would drive fans. They didn't believe the demand was great enough to justify the extra couple of dollars it would add to the price of the board.

Today, however, several manufacturers are implementing mainboard fan speed control on all but their cheapest products. They're doing this for the same reason they're piling on other features, like RAID and WiFi - because processing power alone is no longer enough to sell a mainboard. Asus dubs this feature Q-fan, for Aopen it is SilentTek, for MSI, CoreCell, and Abit's version is called Fan EQ. In addition, just about every small form factor PC on the market uses automatic fan speed control.

I tried out the Asus P5GDC-V Deluxe, a mainboard which can control the speed of the CPU fan and one other fan, based on the CPU temperature. To cool the 3.4GHz P4 CPU, I tested two heatsink fans from Coolermaster, the rather large and heavy Hyper 48, and the simply gargantuan Hyper 6+. Both of these fans are new or updated products, which use the new style 4-pin connection to the motherboard, not the older 3-pin type.

Ultimately, I did most of my testing with the Hyper 6+, that's partly because it ran a bit quieter and cooler than the Hyper 48. However, I also had some problems fitting the Hyper 48 onto the mainboard because of Asus's Stack Cool technology, an extra layer of PCB under the CPU section of the mainboard, which purports to help prevent overheating. Because Stack Cool effectively makes the mainboard several millimeters thicker than is normal, I was very concerned that I could overtighten the bolts holding on the heatsink, and damage the motherboard or CPU. I can't say for certain that these fears are justified - and I have heard of no cases of users actually damaging a Stack Cool motherboard in this way (neither have Asus, Coolermaster, or Thermaltake representatives I spoke with). The Hyper 48 mounting system is also quite rigid, and if you don't fit it very carefully, it's easy to over- or under-tighten one side, causing it to make poor contact with the top of the CPU - my concerns about Stack Cool exacerbated this issue. The Hyper 6+ mounting system includes soft pads, which help spread out the force equally.

Once the CPU fan is fitted, Asus's Q-Fan2 fan control technology is simple to use. You set a maximum temperature in the BIOS setup screen, and then the CPU fan spins fast enough to keep the CPU temperature below it. To prevent the fan spinning so slowly that it actually stops, you also set a minimum fan speed. The fan will not go slower than this and therein lies a problem with Q-Fan. For all the fans I tested, the minimum speed permitted appeared to be too high. For example, the speed of a 3600 RPM fan could not be reduced below 3000 RPM - still quite noisy.

A workaround for this problem is to disconnect the PWM (Pulse Width Modulation) control wire of the fan. This converts a four pin fan to an old style three pin fan. There are various ways to do this. I choose to bend back the PWM pin on the mainboard CPU fan connector so that it would no longer go into the plug (if you try this yourself, there's no need to bend it flat - about 30 degrees from vertical is enough). I also needed to tell the BIOS that I'd done this, by changing the Q-Fan mode setting from PWM to DC. In DC mode, Q-Fan changes the fan speed by varying the fan voltage, rather than sending a PWM signal.

With this modification, Q-Fan functioned perfectly. Most of the time (for example, browsing the Internet, or when the system was playing a movie), the fan speed stayed below 1000 RPM. With both the Hyper 48 and the Hyper 6+, this meant it was difficult to hear the fan at all. Under heavy load - during a game of Half Life 2, for example - the speed would go up to around 1500-1800 RPM, which is audible, but still quite unobtrusive.

I used Panopsys Throttlewatch to help configure Q-Fan. This allowed me to set the CPU fan speed just high enough to prevent the CPU's automatic TM1 throttling from engaging (at somewhere around 70 degrees C), so I could be assured of maximum performance with the minimum of fan noise.

I informed Asus of the fan speed problems with four pin fans, and they have told me that they are working on an updated BIOS which will allow four pin fans to run more slowly - I hope they will implement this improvement on all mainboard models which use Q-Fan2. In the interim, it's also possible to use other software, for example Almico's SpeedFan, to reduce the speed lower than Q-Fan can. However, both the four pin PWM fans I tested seem to have a fixed lower speed limit, around 1500 RPM. There was no way to reduce the speed below this, other than modifying the connection to three pins, as described above.

With the CPU fan noise under control, I turned my attention to other system components. I didn't want to have to use an extra fan to push hot air out of the case, so it was important to try to keep the heat output as low as possible. I installed a pair of Kingmax DDR-II 533 512MB RAM modules. DDR-II uses a lower voltage and runs much cooler than normal DDR at high speeds. It also seems very likely that installing more RAM will decrease system noise and temperature by reducing disk activity (because disk data will tend to be cached in RAM more often) - hence the decision to go with 1GB of RAM.

For graphics, I used GeCube's fanless ATI X700 SilenCool card (see separate review here). Several readers have pointed out that there are a couple of more powerful fanless graphics cards on the market: for example, Gigabyte Technology's heatpipe cooled 6800 card. The X700 SilenCool produced no noise, of course, but it did pump out a tremendous amount of heat, which undoubtedly pushed up the system temperature and caused the CPU and PSU fan to run a little faster. With the graphics card removed (the P5GDC-V includes onboard graphics, by dint of its Intel 915G chipset), the CPU fan and PSU fan both ran roughly 100-300 RPM slower when under maximum load.

Power was supplied by the Coolermaster Real Power 550W. The Real Power 550W was also the sole source of active case ventilation, with its temperature-controlled 12 cm fan.

The Real Power 550W was also extremely quiet. The fan speed remained around its minimum, 1200 RPM, most of the time, only rising to a peak of around 2000 RPM (measured with an optical tachometer, as this model has no fan speed sensor connection) when the system had been under heavy load for twenty minutes or more. This increase in speed was entirely due to the hot air being pulled out of the case by the PSU fan (when running outside a case, the PSU fan speed never changed at all). The power consumption meter included with the PSU indicated that the total power drawn by the system never exceeded 200 Watts. I should note that a 550W PSU is definitely overkill for a system of this nature, a good 400W (or perhaps even 350W) supply should be plenty.

Of the noise sources inside a PC, the hard disk drive is usually the quietest. I simply used soft rubber washers to help prevent the 160GB Western Digital S-ATA drive from transferring low frequency vibrations to the case - I found the spindle noise and seek noise were almost inaudible. Optical drives present a more serious problem. While the platters inside a hard disk drive are manufactured with great precision, and as a result, are perfectly balanced - CD and DVD discs are not. What's more, with the outer edge of the disc exceeding 200 km/h, they spin considerably faster than most hard disk drives. A slightly wonky plastic disc spinning at 12,000 RPM makes a great deal of noise. Turbulent air flow around the spinning disc generates more noise, and the relatively heavy pick up head also creates seek noise as it moves back and forth.

Optical drive manufacturers are beginning to take the noise problem seriously. QuieTrack is Asus's answer, a collection of technologies which reduce vibration and control airflow. Unfortunately, Asus has still not released a QuieTrack DVD writer. I tried out two older QuieTrack drives, the CRW-5232A CD-RW, and the DVD-E616P2 DVD-ROM. The results were mixed. They are the quietest optical drives I've ever heard, but in a very quiet PC, a CD spinning at 12,000 RPM is still the noisiest component. To really get the noise under control, I installed Nero DriveSpeed, a freeware application that lets you set maximum speeds for optical drives. With the maximum set to 24X, the drives were much, much quieter, though still reasonably fast. The only time the lower speed became really noticeable was while installing large applications from CD, which took something like 30 to 50% longer. µ