You are here

SPCR's Fan Testing Methodology [2006]

November 3, 2006 by Devon
Cooke

Fans are by far the most common and problematic sources of noise in computers.
Once quiet fans are found, other noise sources like hard drives, whiny electronic
components, and rattling case panels can be addressed, but a quiet system always
starts with quiet fans. It is therefore a bit surprising that Silent PC Review
has never reviewed fans. There is plenty of good
information to be found in our forums (Felger Carbon
get special mention here), and we do have a Recommended Fans page, but, believe it or not, SPCR has never actually reviewed
a fan. Examined, listened to, pulled apart, measured, and recommended —
many a time — but never reviewed.

Why not? There are many reasons, but here's a big one: Fans aren't really
all that interesting
. The technical details that differentiate fan designs
are complex and difficult to understand, and, when it comes right down to it,
they often don't have much impact anyway. Besides, most of SPCR's visitors really
only want to know one thing when it comes to fans: Which one is quietest?

Answering that question is certainly worthwhile, but it's not a simple question
by any means, and it can't be answered by looking at fans in isolation. A large
roundup is required, and SPCR took the first baby steps in this direction in
early 2004, when we announced a project called Calling
All Good Fans
, with the intent of having a published piece by September
2004.

Two years later, we've amassed hundreds of fans for the project, but we still
have yet to publish anything, while our list of recommended fans has languished
for almost as long. That's about to change, as we have finally amassed the proper
tools, experimented with various measurement techniques, and set aside a large
block of time so that we finish this project properly. This article explains
the hows and whys of fan noise — how we measure fan noise, and why we measure
it the way we do. The actual test data from our collection of fans
will appear periodically over the next few months in batches of twelve or so at
a time.



Our Calling All Good Fans project brought in hundreds of fans from
all over the world.

WHAT MAKES A QUIET FAN QUIET?

Just about any fan can be made quiet if its speed is reduced enough. All other
things being equal, a slower fan is a quieter fan. Keeping this in mind, it
makes sense to begin the search for a quiet fan by examining the various low-speed
fans that are available. But things aren't quite so simple. Most obviously:
Slower, quieter fans don't cool as well as faster, noisier ones. On top of that,
nobody with an interest in noise is likely to run all of their fans at full
speed, so the rated speed is not important. What matters is the actual
speed, and that is up to the end user.

In real life, every system requires some minimum amount of airflow to be cooled
adequately, so there is a practical limit to how slowly the system's fans can
spin. How slowly? That depends on the fan: Some fans can generate more airflow
at a given speed than others. Given what we know about noise and rotation speed,
the quietest fans should be the ones with the highest airflow to speed ratio.

As a general rule, this might be true, but a search for the fan with the best
airflow to speed ratio won't always find the quietest fan. Remember, a slower
fan is a quieter fan all other things being equal. It doesn't take much
time playing with fans to learn that all other things are not equal. What we
really want is the fan with the highest airflow to noise ratio. When
choosing a quiet fan, it turns out that speed is irrelevant. Speed is
a decent guideline for judging how noisy a fan is compared to other fans at
different speeds or for judging how much airflow it produces, but it doesn't
tell us what we really want to know: How much noise does a fan make while
still keeping my system cool?

Airflow to noise ratio is certainly the most important criterion for choosing
a quiet fan, but it is by no means the only one. Once the fan with the best
airflow to noise ratio is found, its speed still needs to be adjusted to provide
the right amount of airflow. If this speed can't be attained, the fan is useless
because it can't provide the correct amount of airflow. Not all fans will start
reliably at low speeds and some fans require very low voltages to run slowly.
A fan that pushes the right amount of airflow at 4V is useless to most users
because very few fan controllers supply less than 5V.

Concerns other than quiet are also relevant, the main ones being reliability,
sample variation, price, and ready availability. Technical features such as
bearing type can affect reliability, and odd design quirks like closed flanges
(which can be incompatible with clip-based mounting systems on heatsinks) or unusual plugs
also need to be taken into account.

HOW WE MEASURE FANS

An SPCR fan test starts with a web search of the fan's model number. If we can't find
a record of it on the web, the fan is set aside — a fan that can't be found
online is not as worthy of a review. Chances are you won't be able to buy one.

DC axial fans on the retail market used to come with minimal packaging. They were mostly surplus stock from large OEM orders that ended up in retailers' inventory for use as replacement parts. Some of the fans we test still fall in this category, and they sometimes come supplied only with bare wire leads. In the past five years or so, the retail market for computer fans must have grown a lot, because so many fans are supplied in fancy packaging, complete with screws or other mounting devices, multiple cable terminations, and even variable speed controllers. All of these added features and details are covered in our reviews.

Fan Specifications

Assuming we do find a record of it, we take note of its official specifications,
taken from the original manufacturer's documents if possible. The specifications
are included as a guideline for what to expect, and give us some idea of how
the fans are supposed to perform. Not all specifications are accurate,
however. It is not uncommon to find multiple specifications that conflict
with each other, especially if they are published at different times or by
different companies.

Retail brands that buy their fans from large OEM companies
occasionally quote different specifications than those of the
original manufacturer. This is especially true for noise specifications;
there are numerous known cases of companies publishing fraudulent
or incomplete noise specifications. Even "honest" looking numbers
don't always mean much, since there is no universally agreed upon standard
for measuring fan noise.

SPCR tracks the following specifications:

  • Model Number
  • Bearing Type
  • Power Rating
  • Airflow Rating
  • Rotation Speed
  • Noise Rating
  • Header Type

Bearing Type

Most of these are self explanatory, but bearing type deserves some explanation.
There are countless bearing designs out there, but most fall into one of two
categories: Ball bearing or sleeve bearing. It is generally agreed that ball
bearings tend to be more durable and last longer than sleeve bearings, especially
in high heat situations. For this reason, most CPU heatsinks and power supplies
use ball bearing fans. In addition, ball bearings are less likely to fail
without warning — worn ball bearings generally start to grind when they
wear, giving an audible warning that something is wrong before they fail entirely.
Sleeve bearings, on the other hand, simply seize up when they fail, which
can be catastrophic if not noticed for some time.

That's not to say that sleeve bearings are always worse. There is some evidence
that sleeve bearings tend to sound smoother than ball bearings, especially
after the fan has seen some wear and tear. They are also slightly cheaper
to make, although that difference doesn't seem to mean much on the retail
market. Lastly, sleeve bearings are more shock resistant. Drop a ball bearing
fan, and the pressure point between the ball and the center shaft is likely
to cause damage — one of the balls may develop a flat spot, a chip, etc.
Sleeve bearings don't suffer from this problem, as the center shaft is evenly
supported by the sleeve across its whole length — there are no pressure
points to cause damage.

There are also several unusual bearing types, such as Hypro, FDB, or Maglev
bearings, and countless unnamed variations on ball and sleeve bearings. Bearing
type can significantly affect reliability and how a fan sounds as it ages,
which is why we report it, but keep in mind that the quality of the
bearing can have as much impact as the underlying technology. Unfortunately,
things like reliability and bearing quality are almost impossible to determine
without looking at hundreds if not thousands of samples. This kind of testing
is far beyond our capabilities, so we leave you with bearing type, and let
you draw your own conclusions.

[Editor's Note: If you undervolt the fans the way most quiet PC enthusiasts do, the wear and tear on them is dramatically reduced. It's not unrealistic to expect a good fan to last a PC's useful life. A few fans in some of SPCR's lab systems have been spinning slowly 24/7 for as long as the site has been live. )

Our Tests

After jotting down all of the relevant specifications, the actual testing
begins. Our two main tests are for noise and airflow — the two parameters
necessary to determine the fan with the best airflow to noise ratio. We also
measure rotation speed and power consumption, and we make an attempt to determine
the lowest voltage at which the fan will start reliably.

Each test is duplicated five times, each with the fan running at a different
speed. Four of the tests run the fan at a constant voltage — 5V, 7V,
9V, and 12V — yielding a good idea of how noise and airflow scale with
speed. The fifth test is run at a constant airflow: 10 CFM for 80mm
fans, 15 CFM for 92mm fans, and 25 CFM for 120mm fans.



A custom-built four channel fan controller is used to produce exactly the
right test voltages.

The constant airflow test is important: It gives a common point of comparison
for every fan tested, making it easy to rank the fans based on their airflow
to noise ratio. Whichever fan is quietest in the constant airflow test will
be crowned the quietest fan overall. The extremely low level of airflow that
was chosen for this test reflects our bias: Most fans will be very quiet at
this level while still providing enough airflow to cool a moderately powerful
system. Some fans, the best fans, are completely inaudible from one meter
at this level. Others never quite disappear entirely but are still tolerable,
while the worst fans exhibit a clearly audible noise signature.

The Fan Harness

All measurements are done with the fan mounted in a specially made harness
that ensures that every test is done with the fan in the same orientation
in the same surroundings. The harness is made out of soft foam that almost
eliminates vibration resonance as a source of noise. This is an ideal mounting
system; any real system will transmit some vibration to whatever the fan is
mounted to, potentially causing resonance. In fact, an earlier version of
the harness built with closed cell foam produced an audible hum when used
with some high vibration fans. Did you know that foam could produce noise?
We sure didn't!



A Foam Harness decouples the fan and reduces resonance noise.



The harness is supported by thick closed-cell foam and a plywood base —

an earlier version had a tendency to collapse under the weight of the fan during
measurement.

Sample Variance

Sample variance in fans can be quite dramatic, and we try and test multiple
samples whenever possible. It is usually not difficult to gather two or three
identical models, but this is hardly an appropriate sample size for true variance
testing. Given the time and expense of collecting (and testing)
even five or ten samples of every fan, we have not made an effort to measure
sample variance.

Even so, we collect quiet fans. Odds are good that if a fan is quiet, we've
received samples from multiple sources and can make some assessment of sample
variance. For this reason our policy is: If we have multiple samples, we will
examine them all and report any variances we find.

[Editor's Note: It is my opinion that damage due to shock during transit or handling is the uncontrollable variable — a dice-roll factor — for small fans. It helps to explain why different shipments of samples from what appears to be the same production batch can sound significantly different as groups. Bearing damage can cause more ticking and buzzing, increased broandband noise, and imbalances that cause subtle axial wobble, which translates into increased vibration. These effects may be imperceptible to a casual user or listener, and I can't imagine any supplier who would accept such samples back as defective or damaged goods, but the differences can be heard and are significant if you seek the lowest noise. I believe that shipping/handling damage explains a big part of sample variance, and some of the differences in experience and judgement among careful listeners/users of fans. This seems especially true for ball-bearing fans. There is no way to prove this without a fancy, large scale study; it's my hypothesis based on some 6~7 years of messing around with a whole lot of small DC axial fans.]

Starting Voltage

The test for starting voltage is intended to determine the minimum voltage
required to start the fan reliably, unassisted. This is very important; a fan may be very
quiet below its starting voltage, but if it doesn't start reliably at that
level, it's not safe for use in a system. Starting voltage is the bare minimum
voltage that the fan can be used at — give it any less, and it quite
simply won't turn on. In fact, we recommend adding a few tenths of a volt
to our measurements just to be safe. Sample variance, bearing wear or damage,
dust, and even temperature can all affect the starting voltage, so it's worth
having a safety buffer in case of any mishaps.

The test itself is quite simple, since it's more or less just guess and test.
We find out what voltages the fan doesn't start at, what voltages it does,
and then start the test somewhere in the middle. We then hunt for the lowest
voltage where the fan will start reliably at least eight times in a row, adjusting
the starting voltage up and down in 0.1V increments until we find the starting
voltage.

Ambient temperature can affect starting voltage. Some fans need lower voltage to start when the ambient temperature is higher, and higher voltage when the temperature is lower. This means that if a computer is left turned off overnight in a room that gets cold, if the room is not warmed up in the morning before the computer is turned on, undervolted fans in that computer may not turn on reliably. The temperature during our tests is usually 20~25°C, which is pretty typical room ambient in temperate climates.

Rotation Speed

Rotation speed is measured in RPM (Revolutions per Minute). A laser-based
digital tachometer from Neiko is used to make the measurements. The tool is
accurate to single RPM differences, which is not hard to believe. RPM is perhaps
the simplest aspect of fan performance to measure; it's a simple matter of
counting the number of times a single blade (identified by a small square
of reflective adhesive) passes under the laser of the tachometer in a given
amount of time and multiplying by an appropriate factor. Unfortunately, as
mentioned above, rotation speed is also the least useful aspect of fan performance.
We include it only because it gives us some idea of blade efficiency. Better
blade designs should move more air at a given rotation speed.



This digital tachometer measures RPM accurately enough to detect single RPM
variances in fan speed.

Power Consumption

Power consumption is measured using a
home-made power meter
that was originally built for measuring power consumption
in hard drives. Power consumption is hardly an important fan characteristic,
but we include it here for sake of comparison to the original specifications.
The amount of power consumed by most fans — and the amount of heat that
they generate — is almost always negligible, especially at the low speeds
that quiet computing requires.



Measuring the voltage drop across a 0.2 Ohm resistor gives us an easy way
to calculate power consumption.

Airflow

Airflow is reported in CFM (Cubic Feet per Minute), but the actual measurements
are somewhat complicated. CFM is used because it is the most commonly declared
fan specification, despite the existence of a more "scientific"
metric measure.

First, an inexpensive windmill anemometer is used to measure air speed.
This reports a value in LFM (Linear Feet per Minute) that approximates the
speed of the air coming out of the fan, but says nothing about the total volume
of air moved by the fan. To determine volume, the LFM value is multiplied
by the amount of open area in the fan, measured in square feet. A quick unit
analysis reveals that LFM (ft/m) multiplied by an area (ft²) yields the
correct unit: CFM (ft³/m).

Using the anemometer is no cakewalk, and its margin of error is unlikely
to be much better than ±10%. Subtle changes in measurement conditions,
such as the angle of the instrument or which side is placed against the fan
can affect measurements severely. In fact, it was the need to standardize
a measurement position for the anemometer that prompted the construction of
the fan harness. For our purposes, the anemometer is always held vertically
as illustrated below, and an orange dot affixed to one side of the anemometer
ensures that the air always blows in the same direction.

Even with these precautions, the position of the instrument relative to the
fan can also affect the measurements, so measurements are taken slowly enough
that time can be taken to find the peak air speed, which is then used for
subsequent calculations. Aside from the audio recordings, this is the most
time consuming part of the test.



Peak airspeed is carefully measured with a windmill anemometer.

Determining the open area of the fan requires a bit of basic geometry. The
area of the motor hub (a dead spot through which no air can travel) is subtracted
from the area of the open circle in the frame. The final formula gets quite
complicated, since all of the measurements have to be converted to feet.

The area of the motor hub is measured, not assumed, since its size can vary
quite a bit. A set of digital calipers is used to make an accurate measurement
to two decimal places.



The fan hub is measured using digital calipers.

[Editor's Note: This particular CFM measurement technique was adopted after several others were tried. The main reason for our choice is that the results are usually repeatable, and they are the closest to the CFM specifications provided for fans from manufacturers whose technical documentation appears reasonably trustworthy. All the other methods gave us considerably lower CFM numbers.]

Noise SPL Measurements

Sound pressure level (SPL) measurements are done from a distance of one meter using our usual
high sensitivity sound level meter: A Bruel & Kjaer (B&K) model 2203.
This professional caliber SLM is >20 years old, weighs over 10 pounds,
and is completely analog in design. It has a dynamic range that spans 140
dB. The unit can measure accurately down to about 16 dBA. A quiet environment
is a prerequisite to low noise testing; the lab has been measured down to
~17 dBA at night.



B&K model 2203 sound level meter measures down to ~16 dBA.

Good as our SLM is, it is not our most important tool for evaluating noise.
While it does a good job of sorting out the coarse differences between fans,
the human ear is a much more sensitive instrument for judging fine nuances.
This is because what we perceive as quiet is much more than simply
low volume. Our recent article called
What is a "Silent" Computer?
describes in detail what we mean
by this, but in a nutshell the quality of noise can be just as important
as the amount.

Don't believe us? Most of the fans that we test measure below the ambient
noise in the constant airflow test. Nevertheless, we still consider some of
them too loud for use in a quiet system. This is not because they are producing
a lot of noise; it's because they're producing the wrong type. The noise of some
fans blend into the background much more easily than others, and these are
the ones that we're looking for.

Audio Recordings

But don't just take our word for it. Try downloading these two sound files:
Smooth Sample, Ticking
Sample
. Each sample contain five samples of ambient noise followed
by five seconds of a fan blowing 10 CFM. Both samples measured less than 18
[email protected] — the ambient noise at the time of recording. The recordings were
made from a distance of one meter and mirror closely what we heard from that
distance.

The two recordings illustrate exactly why sound quality matters: The Smooth
Sample is completely inaudible; you can't tell that there's a fan on. If you
listen very closely, you might hear a slight pop in the middle of the file
that marks where the recording of the fan was joined to the recording of the
ambient noise, but the subsequent noise just sounds like... more ambient.
The Ticking Sample, on the other hand, leaves no ambiguity that there is something
being recorded. You notice when the transition between ambient and
fan noise occurs because there is an audible difference. There's no doubt
that the Smooth Sample is quieter, even though their measured SPL is the same.

Because we test fans at a constant airflow, noise measurements are not that
useful because keeping airflow constant means that most fans measure very
similarly. Instead, we use our ears to determine which fans are quietest...
and we encourage you to use your ears as well.

Every fan that we test is recorded four times, according to our
standard Audio Recording techniques.
The four recordings are as follows:

  1. Alternating ambient noise and the fan running at 5V, 7V, 9V, and 12V,
    recorded at a distance of one meter.
  2. Alternating ambient noise and the fan running at 5V, 7V, 9V, and 12V,
    recorded at a distance of one foot (30 cm).
  3. Five seconds of ambient noise, followed by the fan running in the constant
    airflow test, recorded at a distance of one meter.
  4. Five seconds of ambient noise, followed by the fan running in the constant
    airflow test, recorded at a distance of one foot (30 cm).

As always, we recommend that you listen and compare the recordings in a specific
way. The green box below describes how we make our recordings and what you're
supposed to do with them.

HOW TO LISTEN & COMPARE

These recordings were
made with a high resolution, studio quality, digital recording system,
then converted to LAME 128kbps encoded MP3s. We've listened long and
hard to ensure there is no audible degradation from the original WAV
files to these MP3s. They represent a quick snapshot of what we heard
during the review. Two recordings of each noise level were made, one
from a distance of one meter, and another from one
foot
away.

The one meter recording
is intended to give you an idea of how the subject of this review
sound in actual use — one meter is a reasonable typical distance
between a computer or computer component and your ear. The recording
contains stretches of ambient noise that you can use to judge the
relative loudness of the subject. For best results, set your volume
control so that the ambient noise is just barely audible. Be aware
that very quiet subjects may not be audible — if we couldn't
hear it from one meter, chances are we couldn't record it either!

The one foot recording
is designed to bring out the fine details of the noise. Use this recording
with caution! Although more detailed, it may not represent how the
subject sounds in actual use. It is best to listen to this recording
after you have listened to the one meter recording.

More details about how
we make these recordings can be found in our short article: Audio
Recording Methods Revised
.

Vibrations

Just as with hard drives, the direct airborne noise is not the only component of a fan's acoustic signature. Fans also vibrate, and those vibrations can cause audible noise. The balance and precision of the bearings, the way the the blades interact with the air, the composition and structure of the frame — all of these can affect the vibrations exhibited by a fan.

To examine fan vibrations, we've adopted the technique we developed originally for hard drive testing. We gauge the effects of vibration by placing the fan on an aluminum electronics project box (43cm x 25cm x 10cm). The box acts as a sounding board and resonating chamber for the fan
vibrations. The box exaggerates the vibration-induced noise to make it easier for
us to hear and evaluate. The air inside the box resonates along with the
thin aluminum panels at the primary RPM frequency and
its harmonics, and any resonant frequencies intrinsic to the material and structure. Some plastics have a pronounced resonance at certain frequencies that are easily excited at specific RPM speeds, but not at others.



The aluminum box amplifies vibrations so we can hear and assess them more easily.

Because of the time and effort required, we will limit vibration testing to those fans whose airborne acoustics are already judged to be good. It's not yet clear whether we can use the same 1-10 scale used for the hard drives. The quality of the sounds are quite different from that of hard drives due to the much lower mass and rotational speed of the fans, which makes comparison between drives and fans difficult. (Because the spin speed of quiet fans is many times slower than the 5400 or 7200 RPM of hard drives, the fundamental audio frequencies are much lower.) Whether we use the same scale or develop a new one just for fans, we will be commenting on vibration, particularly at the speeds that each fan is most likely to be run by PC silencers.

CONCLUSION

With this methodology in place, SPCR can finally begin to seriously undertake
the project we began more than two years ago. There are plenty of fan round-ups
on the web, but we think you'll agree that none of them covers noise as extensively
or as realistically as we do. We've worked hard to make sure that the fruits
of our labor are as accurate and as useful as possible. We are confident that
we have taken a big step towards filling a sizable gap in the knowledge available
on SPCR, but, as always we are sure there are things we have missed. If you
have suggestions or improvements for SPCR's Fan Methodology v2.0, don't hesitate
to tell
us about them in our forums
!

[Editor's Note: Some readers will be asking, "But what about fan blade geometry and design?!" Well, this is a fascinating topic, but it's unrealistic for us to try and tackle it. For one thing, it is extremely difficult to correlate a fan blade design with airflow or noise performance. Secondly, we don't really care how a fan manages to be quiet, but that it is quiet. Certainly we'll touch upon unusual aspects of a fan's design, including its blade geometry; no attempt will be made to address the performance consequences of those design aspects, however. ]

Google

www SPCR