A new fan test system for 2010 focuses on the desired end result of fans in a PC: Cooling, as seen in temperature drop. Measured airflow ranks low on measured parameters; noise remains priority number one. No one has tested PC fans this way before.
The role of 12V DC axial fans in PCs dates back over thirty years. They’re
the quick and easy cooling tool of choice for engineers, and they’re often the
bane of noise-haters. SPCR
has been studying, testing, analyzing, discussing and thinking about DC axial
fans since our inception. Fans seem unbelievably cheap to produce, they
are made in enormous quantities, and they’re ubiquitous around computers. They
are usually too noisy.
Our first assessments of fans was based almost entirely on listening, and this
method was good enough to identify several fans that were markedly better than
others available at the time. It took some years for us to develop a fan
test methodology that was better than careful listening. We combined subjective
listening, airflow measurements and SPL measurements at a wide variety of speeds
to establish a more complete profile of each fan. The airflow measurement methodology
evolved over several sets of test runs, with the last one being A
New Way to Test Fan Airflow… but we were never fully satisfied with out
results.
Fan manufacturers use complex multi-chamber tools to measure airflow. The cost
of these tools runs into many thousands of dollars, possibly tens of thousands.
The excerpt below from Laboratory Methods of Testing Fans for Aerodynamic
Performance Rating, document ANSI / AMCA STANDARD 210, shows one of
the simplest fan airflow measurement devices illustrated in that document. It’s
no wonder that our self-designed and built airflow measurement systems didn’t
quite hold up.
 |
Manufacturers’ most commonly cited airflow measurement is free-air — that
is, the airflow in cubic volume per minute (Cubic Feet per Minute,
for example) when the fan sees no resistance or impedance to its movement other
than still air, at sea level pressure in moderate temperature and humidity.
CFM always varies with impedance, and "serious" fan manufacturers
measure this under varying impedance, often at different RPM with each fan.
For fan makers, CFM is one of the main selling or defining parameters for a
fan. Thermal engineers have access to details on typical and safe maximum temperature,
impedance to airflow, the heat transfer parameters of heatsinks, and many other
complex factors other factors. From such data, thermal engineers can use heat
transfer and fan equations to calculate airflow requirements, which are then
used to specify appropriate fans.
That is for thermal engineers. For typical PC enthusiasts, CFM is a purely
abstract concept. This is not to say that there’s no relationship between
airflow and cooling, but that there is no recognizably linear relationship,
and it differs for each thermal system. We simply do not know
clearly enough the relationship between the CFM rating of a fan and the cooling
it effects in our thermal system.
This is very different from many other technical issues around computers that
DIY PC enthusiasts deal with. With power, for example, a watt is a watt, whether
it’s in AC or DC, and a watt drawn by the PC means there’s a watt of heat being
dissipated by the PC. If you need to know how much power the components are
using, as opposed to the whole PC, then you apply the AC/DC conversion rate
that applies. For example, a system that draws 200W from the wall and uses a
PSU know to be 80% efficient at that power level is delivering 160W DC to the
components within. This methodology of estimating DC power use remains the same
for any type of PC. There is no similarly neat way to calculate how airflow
and cooling work together.
A simple example: If a fan spinning atop a CPU heatsink effects a temperature
rise of 20°C in the CPU under 100% load, all we can predict is that if the
fan is slowed down, the CPU will run hotter, and if the fan is sped up, the
CPU will run probably run cooler. There is no way for us to predict with
any scientific certainty just how much the airflow must change to effect, say,
a 5°C change in the temperature rise. Furthermore, the amount of airflow
change needed will not be the same going up in temperature compared to going
down. We also cannot change airflow directly; it changes in response to the
RPM, which we can change, but there’s also no precise correlation between RPM
and CFM.
Finally, CFM has no direct bearing on cooling, which is measured
not by airflow but by a drop in temperature, usually in a device
in the PC. Without the benefit of a thermal engineer’s knowledge and detailed
parameters about the components and conditions, CFM might as well be APH (angels
per pin head). The relationship between CFM and cooling is at least as complex
as that between SPL and perceived noise. The CFM value has no real meaning beyond
itself. In contrast, with a bit of experience, 30 dBA/1m does have some meaning.
Still, DIY computer tech geeks want to compare fans by their CFM rating, and
in the SPCR (and other PC tech web) forums, some have gone so far as to specify
what CFM rating they believe is needed for their application. This is a reliance
on CFM numbers that has obfuscated the role of airflow in cooling. It’s not
really a surprise; CFM is one of the very few performance specs that fan manufacturers
make available.
Over the years, we have observed one clear phenomenon about fans and cooling:
The relationship between airflow and temperature invariably becomes exponential at some point. Increase airflow from nothing to something, and the drop in temperature
can be dramatic. Keep increasing airflow, and the cooling improvement becomes
less and less significant, until at some point, the temperature hardly drops
at all. The trick, for the PC builder who seeks both good cooling and low noise,
is to find the point where any decrease in airflow (or fan speed) effects a
significant increase in temperature, while only a very large airflow increase
effects a significant temperature drop. In other words, once you have enough
airflow, additional airflow has very little cooling effect, so all you’re doing
is increasing noise. "Enough airflow" is not a constant, of course,
it varies for each system of components.
 |
When our previous fan testing strategies were reviewed against this brackdrop,
we realized that our attempt to accurately measure CFM was a kind of search
for the holy grail. It simply was not going to be achieved successfully, not
without a dramatic, multi-fold increase in funding and expertise. As is so common,
this realization opened up a new opportunity, a new way of looking at the performance
of fans in the context of silent PC cooling. Many months of planning, thinking
and experimentation later, we have a new fan testing system, one that’s a dramatic
departure from all of our previous systems.
NEW FAN TEST SYSTEM
There are two parts in our fan testing system. One part concerns audio measurements,
the other involves cooling performance.
A. Audio Instrumentation and Resources (fully detailed in New
Audio Test Gear, SPCR 2008) — Our audio measurement tools are now far
more sophisticated than before; it is SPCR’s anechoic chamber and 9dBA-capable
instrumentation. We can assess not only the SPL of a fan at different speeds,
but also its frequency spectrum, so that tonal peaks can be identified and visualized.
- ACO Pacific PS9200KIT
Type 1 Measurement Microphone System - Landtek ND9 Microphone Level Calibrator
- SpectraPLUS
– Audio Spectrum Real Time Analyzer - M-Audio Tampa professional microphone/instrument preamp with integrated
24-bit / 96-kHz A/D converter - M-Audio FireWire 410
- Silent PC for audio analysis and recording
- Anechoic Chamber
ACO Pacific 1" diaphragm capsule mic with adapter on 1/2" preamp.
SpectraPlus screenshots.
M-Audio FireWire 410 atop M-Audio Tampa digital mic preamp.
B. Fan Test Gear — Our latest airflow measurement device (anemometer)
is by far the most accurate of the handful that we’ve acquired over the years.
It is used not as a primary tool, however, but a secondary one. Our approach
this time around is not to be concerned about airflow per se, but its
thermal effects in a cooling system. CFM is relegated back to its original place
when we first began assessing fans years ago: Mostly a point of note. Our
most important measured parameter is temperature rise, not by itself but plotted
against SPL.
- i7-1366 CPU die simulator with embedded T-type Thermocouple wire
- Mastech
6030D DC Regulated Power supply, 0-64V/3A - Custom built 0-12VDC Regulated Fan Controller
- Kanomax 6803 Anemometer
- Mannix DT8852 Dual Input Thermometer (K, J or T Thermocouple input)
- High accuracy general purpose Multimeter
- Laser digital tachometer by Neiko Tools USA
- Calibrated electronic strobe light (alternate tachometer)
 This is our fan testing setup as of May 2010. (Cick on image to enlarge.) |
THE TEST PROCEDURE
The gist of the fan test system is fairly straightforward:
- A CPU simulator, in the form of copper block that has the same dimensions
as an Intel i7-1366, has a small heater coil embedded within, capable of handling
150W. - A large heatsink with somewhat high airflow impedance is mounted to cool
the CPU simulator. - The heater coil is powered by a regulated lab power supply to 137W. (Typically,
64.6VDC x 2.1A). This is the maximum power that the heater coil can pull from
the lab power supply, although there is a bit of headroom in the coil as wenn
as the PSU, which is rated for maximum 3A at 64V (192W). - The fan to be tested is mounted on the heatsink and driven by a regulated
0~12 VDC power supply at standard voltages, speeds, and SPLs. - The SPL is recorded (in [email protected]) in the anechoic chamber at every voltage
level. - The temperature of the CPU block and that of the air 6" in front of
the fan is monitored closely using T-type thermocouple wire sensors and a
dual-input digital thermometer. - A precision anemometer is used to record air velocity (Feet Per Minute)
at every speed and SPL. - The most important parameter is Temperature Rise vs SPL.
We refer so often to temperature rise at SPCR that we sometimes forget
that not everyone lives and breathes it. Basically, it refers to the difference
between ambient temperature and the temperature of an object under themal load.
Better cooling results in lower temperature rise; worse cooling results in higher
temperature rise. In this case, the ambient is the temperature of the air 6"
in front of the fan, and the thermal load temperature is that of the CPU die
simulator.
In the past, we’ve used 12, 9, 7 and 5 volt settings for the fan drive voltage
as test points. It made sense for a long time, as these voltages are fairly
easy to obtain in any PC (except for 9V). Today, there are many more ways to
adjust fan speed. Most motherboards are equipped with speed controllers for
their fan headers, and monitor fan speeds for any standard 3-pin fans or 4-pin
PWM fans. In most cases, the RPM can be displayed right on the desktop using
any number of fan / thermal utilities.. So now, we’re using specific RPM for
the primary test points. Since we have little reason to change our long-standing
reference of the Nexus 120 fan, its RPM at 12, 9, 7 and 5 volts will be used
for standard test points. Above the 1080 RPM maximum speed of the Nexus 120,
we will choose test points based on the performance of other, faster fans.
A PEEK AT EARLY RESULTS
The assessment of a fan is always comparative; performance can only be judged
in comparison with other fans. There are a small number of fans that we’ve been
using routinely in our lab due mostly to their low overall noise and benign
sonic qualities. Here is a summary of the test data for a few fans, obtained
from our new test system.
Astute readers will note that the temperature rise figures obtained here are
lower than with the i7 CPU on our current heatsink test platform, despite the
similar 137W load. This is to be expected as the entire copper top of the die
simulator is radiating heat more evenly than a real i7; hence it runs a bit
cooler. This is not particularly important in our fan testing procedure, however.
The instant, easy, repeatability of the power settings is far more important.
Each fan was tested at maximum speed (12V), and then at voltages that provided
1080, 860, and 720 RPM. These are the speeds that the Nexus 120 achieves at
12, 9, and 7V. The high speed Scythe Ultra Kzae fan could only be tested at
one speed to match the Nexus 120, 1060 RPM at 4.6V. It would not start reliably
much below that voltage.
| 120mm Fans on Thermalright U120E + 137W Thermal Load | ||||
| RPM | VDC | FPM* | SPL ([email protected]) | °C RISE |
| Nexus 120 (reference fan) | ||||
| 1080 | 12 | 240 | 16 | 21.0 |
| 860 | 9 | 180 | 13 | 24.2 |
| 720 | 7 | 150 | 11 | 29.2 |
| 550 | 5 | 100 | <10 | 38.2 |
| Scythe Slipstream SY1225SL12M (medium speed) | ||||
| 1390 | 12 | 370 | 21 | 18.7 |
| 1080 | 7.8 | 285 | 16 | 21.6 |
| 860 | 5.4 | 270 | 13 | 25.2 |
| 720 | 4.3 | 150 | 11 | 29.0 |
| Noctua NF-S12-1200 (original design) | ||||
| 1430 | 12 | 420 | 21 | 20.7 |
| 1080 | 8.4 | 325 | 16 | 23.9 |
| 860 | 6.3 | 285 | 13 | 28.0 |
| 720 | 5.1 | 220 | 11 | 32.3 |
| Scythe Ultra Kaze DFS1238H-3000 (120x38mm) | ||||
| 2800 | 11.9 | 650 | 44 | 14.8 |
| 2280 | 9 | 530 | 38 | 15.5 |
| 1800 | 7 | 420 | 32 | 17.0 |
| 1230 | 5 | 270 | 25 | 21.0 |
| 1080 | 4.6 | 240 | 22 | 22.5 |
| *FPM = Feet Per Minute. This is the actual value that an anemometer measures, the velocity of the airflow through its vane. The widely used CFM is Cubic Feet per Minute, obtained by multiplying FPM by the area of the inlet or exhaust. In previous reviews, we measured FPM directly at the fan, and multiplied that value by the area of the fan blades (area of diameter minus area of center hub). This was always a bit of of a scientific guess; no more guessing. The FPM is provided, and we don’t believe it differs much from CFM for fans of the same diameter. In other words, our FPM measurements can be compared much like CFM, if you feel this is important. We caution you, however, that like CFM, FPM does not correlate that closely with temperature rise. | ||||
For anyone interested in the relationship between RPM, air speed, and cooling,
the above table is fascinating. Here are some observations.
1) Nexus 120 remains an excellent choice for a quiet reference fan. It
is quieter than the other fans at the 1080rpm max speed, and its cooling is a
half degree better than the Slipstream, 3°C better than the Noctua and
2.5°C better than the Ultra Kaze. At 860rpm and 13 [email protected], its advantage
increases over the Slipstream and Noctua, to 1°C and 4°C. Its overall
noise signature is more pleasant than the Noctua, and a bit of a tosseup against
the Slipstream, though if pressed, I’d choose the Nexus.2) Each fan has a somewhat different relationship between RPM, FPM and
cooling. The Nexus measures the lowest FPM at any given RPM, while the
Noctua measures highest. It’s obvious that lower RPM leads to lower FPM. So
reduce the speed of a fan, and the cooling suffers. But there
is no logical correlation between FPM and temperature rise from one
fan to another. For example, 180 FPM on the Nexus gave 24.2°C,
a degree better than 270 FPM on the Slipstream. On the other hand, an even
higher 285 FPM on the Noctua gave the much worse result of 28°C.3) Much higher fan speed does not mean proportionately better cooling.
The Sycthe Ultra Kaze is a 38mm thick fan rated at 3000 RPM. It didn’t quite
reach rated speed in our test rig, but note the differences in cooling at
the top three speeds — they are separated by little more than two degrees,
while the noise spread is a massive 12 dBA! At 7V, 1800 RPM and 420 FPM, we
saw 17°C temperature rise with 32 [email protected] SPL; increase the speed a thousand RPM and the noise
by 12 dBA (to a whopping 44 dBA) and the cooling improved by only 2.2°C. This suggests that for
the thermal load of our test rig, the relationship between 120mm fan speed
and cooling becomes exponential around the 2300 RPM mark. Increasing fan speed
beyong that point provides only marginal increases; there’s only 1.3°C
improvement going from 2280 to 2800 RPM.Interestingly, when the powerful 38mm Scythe is set to the 1080rpm reference
speed, the temperature goes 1.5°C higher than the Nexus, while
the noise level is 6 dBA higher — with a terrible, tonal quality in stark
contrast to the smoothness of the Nexus. Certainly, there seems to be no pressure
advantage in the deeper 38mm blades compared to the standard 25mm ones at
this speed. Comparing the best cooling results of the Nexus and the Ultra
Kaze, you have to ask if anyone would be willing to accept a 28 [email protected]
noise penalty for 6°C cooling improvement.4) With a lower impedance (less tightly spaced fins) heatsink of similar
quality and size as the U120E, less airflow will be needed for the same
results, In other words, the lead of the Nexus 120 at low speeds will naturally
increase. Tighter fins spacing in the same size heatsink will mean a higher
number of fins and greater fins surface area, so a converse result with this
hypothetical heatsink is that higher airflow will probably provide better
cooling. Most readers should have little interest in a heatsink with tighter
fin spacing than the U120E, as it will require a faster, noisier fan to reach
the same cooling, even though its ultimate cooling capability could be higher
(due to the combination of higher fin surface area and higher airflow).
Suitability of the Thermalright Ultra 120 Extreme as a General Purpose
Thermal Load for Fan Testing
The Ultra 120 Extreme was not just randomly chosen. It was one of four heatsinks
considered. We think it is a good balance between high and low impedance.
Fans are used directly on heatsinks, and as exhaust or intake case fans. On
a heatsink, a fan might face a higher impedance than as a case fan. In either
case, the thermal loads which a fan is asked to cool are generally equipped
with heatsinks (CPU, VGA, NB, VRM), and there is always some impedance.
The other heatsinks considered… and rejected:
- Prolimatech Megahalems – It was our first choice, as its mounting
systems was the best, its fin spacing seemed about right, and its fan clips
are very easy to use. After many days of fan testing and examining the results,
we came to the conclusion that that the Megahalems is simply too good with
low airflow; it shows too little difference in temperature between low and
very low speeds. It also doesn’t show much cooling improvement beyond about
1600~1800 RPM, which is a bit low considering the range of fans we have to
test. The ideal setup is one that shows the greatest range of temperature
differences for different fan speed/airflow settings. - Scythe Mugen 2 – Fin spacing and impedance appropriate but mounting
incompatible w/ thermal test platform. - Noctua NH-D14 – Impedance about right, but load on both sides of
fan (in middle position) causes unnaturally high turbulence noise with higher
speed fans.
In any case, the test results do not turn out that differently. The
Thermalright Ultra 120 Extreme has higher impedance due to its tighter fin spacing
(1.5mm vs 2mm in the Megahalems), and the Megahalems also has the gap in the
biddle between its two banks of fins for even lower impedance. On the U120E,
the Nexus 120 shows a wider range of temperature drops at different speeds,
with some 17°C between 1080 and 550 RPM, compared to 14°C on the Megahalems.
The Megahalems also cools a bit better, especially at the lower speeds. But
just as on the Ultra 120 Extreme, the 38mm high speed Ultra Kaze fared 1.5°C
worse at the same 1080 RPM as the Nexus. The data basically shows the slower
Nexus faring less well at slower speeds on the Thermalright than on the Megahalems,
due to the former’s higher impedance. The U120E shows greater differences in
temperatures over a wider range of fan speeds than the Megahalems, so it is
better choice for general fan performance testing.
Noteworthy points on the New Fan Test System
1) Using the die simulator and DC power supply is an incredible luxury compared
to the CPU, heatsink, or fan testing we’re used to doing all these years. You
turn on the power supply, dial up the needed voltage and current, turn on the
fan, and within 5 minutes you have a perfectly stable temperature reading. There
are no mysterious software or CPU-based fluctuations in either power or temperature;
they stay stable within a watt or a degree. Getting the fans to stabilize at
the exact RPM we seeks is tricky, but we usually get it within 10 RPM. Keep
in mind that with higher ambient temperature, the fans seem to run just a touch
faster at the same voltage.
2) Testing fans smaller or larger than 120 mm diameter will require choosing
other reference heatsinks, We already have a good candidate for 140 mm fans,
the Prolimatech Armageddon. We’ll choose an appropriate one for 80/92 mm fans
when the need comes. Ditto smaller fans.
SUMMARY
We are reversing several years of fan testing by putting airflow measurements
back to the very bottom of relevant parameters that we seek to measure. Those
who have been reading SPCR articles and forums from the start know that this
was our original position, and we were swayed to look more closely at airflow
by popular opinion. Instead, we’re focused now on the end result of airflow:
Cooling, by comparing the temperature drop of each fan at specific speeds and
voltages on a well-known and extensively examined heatsink with a static, consistent
137W heat source on a block of copper that emulates an Intel i7-1366 processor.
Some will argue that while this approach tells about the performance of a fan
when mounted on a heatsinks, it tells much less about when the fan is used for
case intake or exhaust. Our counter is that even in the role of a case fan,
the main question is how much cooling effect the fan has on hot devices that
generally have heatsinks mounted on them. Unless the temperature difference
between the air outside the case and inside is huge, the relationship between
airflow and temperature drop follows the same type of curve shown on the first
page of this article; beyond a certain point, additional airflow becomes essentially
useless. Precise setting of airflow for case cooling is also less critical than
with fans for use on heatsinks, simply because most hot components in computers
do have heatsinks and fans on them.
The best way to determine how much case airflow is needed for optimum cooling
is to experiment with the system you wish to cool, under realistic conditions.
There’s usually a subjective/objective balancing by each DIY PC enthusiast between
spot-cooling fans (such as the fans on CPU heatsink or GPU heatsink, or in a
PSU) and case fans. What works best for one user and system doesn’t necessarily
work for another. Our reductionist point of view is that for a silent PC, what
is easily perceived but very difficult to change is the fundamental sound character
of a fan, so it’s always best to choose a fan that sounds best and work with
it (or them) to achieve the cooling you need. Look for fan reviews using our
new fan testing system in the near future.
* * *
Articles of related interest:
A New Way of Testing Fan Airflow
SPCR’s Fan Testing Methodology
Anatomy of the Silent Fan
Recommended Fans
* * *
