Building a Silent PC: My Quest for Quiet

Chapter 2: Underclocker's Anonymous

Terry Gray



Original: 11 Oct 2000
Revised: Apr 2002
Revised: 14 Jul 2003

Building a fan-free Pentium computer: Design Issues

The PC noise problem is a three-legged stool: power-supply fans, CPU fans, and hard-drives. (OK, recent high-end video cards also have fans, but I'm not a gamer, so I can ignore that problem for a little while longer.) To date I have successfully removed two of the three legs: I have built a couple of PCs that use convection-cooled power-supplies and CPUs. I'm still working on the hard drive problem, but even the current solution is infinitely more pleasant than any commercially-available PC that I know about. (I've heard that some ultra-quiet PCs are being built in Europe, but I have no first-hand experience with them.)

Power Supply Issues

Finding a convection-cooled ATX power supply, complete with standard cables and connectors, is a challenge. TKPower in California makes some, but they don't (as of this writing) mention them on their website (www.tkpower.com) so you have to ask. [Update: TKpower apparently no longer sells to individuals, but this power supply is currently available from Silicon Acoustics in Washington state.] These are not exact physical replacements for regular PC ATX power supplies, they are open frame designs intended for low-profile (1U) rack-mount servers. So a little mechanical work is required to mount them. They are also rather spendy, compared to normal PC supplies. [Update 2: (Jul 03) The TKpower unit has proven unsuitable for P4 systems; however, Silicon Acoustics now carries the SilentMaxx proSilence PCS-350 which is a completely fan-free ATX-12 supply in a standard ATX form factor. It's spendy, but as close to the ultimate PC power supply as I've seen.]

A burning question (pun intended) is whether an over-sized standard power supply can be used for light duty without forced-air cooling. For example, a 300 or 400 watt supply used in a system that only requires 50-75 watts might give long life even without its cooling fan. I *suspect* this is true, but haven't fully explored the issues. And I doubt that I could get a straight answer from the power supply vendors, since they would no doubt be paranoid about liability issues if they said it was OK for a customer to take their p/s apart to remove the fan.

Processor Issues

While there is a huge literature on the web concerning overclocking processors to get the last drop of performance out of them, there is hardly anything written about underclocking CPUs, which is --along with an excellent heat sink-- the key to running CPUs without forced-air cooling; i.e. fan-free. With a really superb heatsink and thermal transfer compound, it may be possible to run some modern processors at full speed without fan cooling, but let's be conservative and consider the underclocking case first. Actually, undervolting is the real key to lowering heat output of a CPU, but reducing the core voltage usually requires running the CPU at a lower frequency, hence we'll just use the term underclocking to imply reducing both voltage and frequency.

A note on manufacturing process rule size. The current generation of CPUs utilize .18 micron design rules, with .13 micron processors expected in 2001. Smaller numbers imply that less voltage is required, and therefore less power consumption for a given CPU speed. Of course, the main driver for smaller design rules is to allow the CPU speed to go up, which results in an offsetting increase in power dissipation, but this is definitely an area where size matters, and smaller is better.

Although AMD is currently providing better price-performance than Intel especially in the economy class of processors (Duron vs. Celeron), AMD processors tend to consume quite a bit more power than Intel CPUs of the same clock speed. Accordingly, Intel CPUs are a better bet for running without a fan.

Modern Intel and AMD processors are "multiplier locked". This means that the processor will only operate at a speed which is a specific multiplier of Front-Side Bus frequency. For example a Pentium III 500/100 has a fixed ratio of 5; it will operate at 500MHz with a 100MHz FSB frequency.

The ability to control FSB frequency, and therefore the frequency of the processor, is limited by the design of the motherboard. I have not seen a motherboard that allows setting the FSB below 66MHz, which until recently was the only option. Therefore, to underclock a multiplier-locked processor we need to purchase one designed to run at a FSB speed greater than 66MHz, i.e. either 100MHz or 133MHz.

Because Celeron processors are still not available at FSB speeds above 66MHz, they are unsuitable for underclocking. For my experiments, I selected a Pentium III 667EB. This CPU is designed to use a 133MHz FSB, so it offers two underclocking options: 500/100 and 333/66. I have found that with a good heatsink, it runs comfortably cool at 500/100 without a fan. That means CPU on-chip sensor temperatures are generally under 100F when doing light tasks, and under 123F (50C) even after periods of intense computation.

Here is an excerpt from a note on my original temperature measurements. Note that I subsequently decided that running at 500/100 (rather than 333/66) would be OK, and so it has proven, since I'm currently seeing a CPU (on board sensor) temp of 32C, and rarely have I managed to get to 50C.


 Experience underclocking a P III 667EB

I purchased a Pentium III 667EB to test the hypothesis that underclocking
a .18 micron processor is a viable way to get reasonable performance
without needing a CPU fan.

I chose the 667EB (supports 133MHz FSB speed) because it has a clock
multiplier of 5, which would let me run it at either 500/100 or 333/66.

Here's what I observed, using the BIOS temperature monitor on an ABIT SE6
motherboard:


    SPEED (CPU/FSB)   CORE VOLTAGE   FAN?   CPU TEMP


Trying 3 possible frequencies with fan and std Vcore:

      333/66            1.63         Yes     29C/84F
      500/100           1.63         Yes     32C/89F
      667/133           1.63         Yes     35C/95F


Looking at effect of lowering Vcore:

      333/66            1.55         Yes     27C/80F
      333/66            1.45         Yes     26C/78F
      333/66            1.35         Yes     26C/78F
      333/66            1.29         Yes     26C/77F

Unplugging the CPU fan:

      333/66            1.63         No      55C/131F
      333/66            1.29         No      43C/109F

Next day, with ambient temp a couple of degrees higher (72F):

      333/66            1.29         No      44C/111F
      500/100           1.29         No      55C/131F


I was a little surprised that the system booted fine at 500/100 with Vcore
set to 1.30 (measured at 1.29).  But I decided that the 500/100 no-fan
temps were a bit higher than I was comfortable with, so I went back to
running at 333/66.

In normal operation, I'm seeing CPU temps in the low 30s (C), but that's
just web brosing and email; I haven't done any extensive testing with CPU
intensive apps such as multimedia or SETI@Home.  I don't know how hard the
BIOS monitor program pushes the CPU, so I don't assume that the measured
temps above are maxima.

Mass Storage Issues

There's more to be said here than I have time to cover now... but in the interim:
Quantum LCT15 drives are pretty quiet, though slow. Fujitsu MPF series drives are not too bad, though I'm told that the 7200 RPM variants are actually quieter than the 5400 RPM units I have tried because they use "fluid bearings" in the higher speed version. However, I'm currently waiting to get my hands on one of Fujitsu's new MPG series drives, which apparently use and even more advanced fluid bearing design throughout that family.

Regardless of how quiet the drive is, the best case is to not have it spinning... hence the importance of RAMDISK and aggressive spin-down policies. Note that there are some serious tradeoffs here. Aggressive use of RAMDISK (while spendy) can speed a system up, whereas aggressive disk spin-down policies will slow the system down whenever disk I/O is required --and with MS Windows it is required more than you would think, even if you give the OS essentially infinite amounts of RAM to use. Oh well.

Current Configuration

My current quiet system configuration is as follows:

I generally use Windows 2000, but can't think of any reason I would change the hardware config for Win 9x or Linux.

Details/Approximate Hardware Costs


This is one of a series of articles describing my silent computing adventures. This link will take you to the beginning of the story.

TEG HOME