A couple comments:

Experimental design #2 could have been modified to include multiple anemometers to reduct airflow restriction for higher airflow fans. For example, had the test rig been modified to hold 4 anemometers, then you simply add the results of all 4 meters to get your airflow reading. This would also provide a way for you to easily change the resistance of the box to see how that affects airflow by measuring airflow with 1, 2, 3 and 4 anemometers.

The hotwire anemometer used in Experimental design #3 looks just like your typical mass airflow sensor that is just about every single modern fuel injected car on the market today.

They are also notorious for being extremely susceptible to turbulence which will affect the airflow readings which can change airflow across the area of the tube the MAF is reading air through.

Typically, they are mounted in an area where airflow will be as smooth as possible, and even then often the manufacturer will build in some sort of "airflow straightener" usually consisting of a hexagonal grid or a simple wire mesh in front of the sensor.

Also, obviously the the area of the opening the hot-wire anemometer is placed in will directly affect the airflow readings unless your meter has some magical way of accounting for the area of the opening, so I assume that you can plug in the area of the opening as a setting into the anemometer.

I would suggest that you build some sort of box or tube to house the anemometer which should reduce turbulence around the anemometer and also provide more reliable measurements. You could easily build a paperboard box with scissors and tape to be inserted into the PSU hole and also fix a grid at the entrance to help direct airflow more smoothly through the box. That should be very easy to build and test.

mutiple anemometers -- impractical & costly. We don't WANT to vary the impedance of the box.

AFA we can tell, both of our anemometers are sensitive to turbulence. However, there isn't much turbulence at the exhaust end after the flow comes around the paper curve, which does help straighten the flow. I don't see the need for any box or tube around the anenometer.

The nature of a spinning fan leads the flow to bounce up/down at least a bit, and we see this reflected in the high/low measurements. The up/down range is quite small (under 5%).

One thing I've been playing with this afternoon is a plastic grid about 2" inside the box, in front of the fan. Each hole is 1.5cm square, and the depth is about 3cm. All of the fan's flow has to go through it. A picture is worth many words here:



Even though it seems to help straighten the flow a bit more, it doesn't do it completely. The measured airflow becomes a bit more even across the exhaust vent -- instead of +10% on the left side, it's less than +5%. The drop on the right side doesn't change though. However, overall measured flow drops about 5%, which means it adds a bit more impedance.

Not convinced that this addition is worth it, as there still seems to be a need to take multiple readings and average them.

BTW.... of course we multiply the area of the exhaust vent with the measured LFM to obtain CFM. There's no other way to do this.

Finally, please take heed of our parting words -- CFM is not that important. Really. If you haven't got that, you haven't really absorbed the article.

Genius


Andy

Awesome article.

You have clearly tested SilenX Ixtrema, any comments on its noise signature?

But wouldn't testing the airflow of the fan under different loads/impedances also be informative? Very often a fan is installed in less than ideal conditions where impedance is high.

Yep, I would expect anything measure airflow at such flow rates to be relatively susceptible to turbulence. I wonder if you could use smoke from a candle or something to help view turbulence through the case.


Yes, I would expect that to help some. Have you also tried it in front of the exhaust vent?

I know the baffle is there to prevent direct airflow into the anemometer, but I have to wonder if the proximity of the of the baffle to the fan and the anemometer might effect readings, especially as airflow goes up.


Just making sure since it wasn't mentioned in the article. Not that it'd matter as long as the relative measurements were consistent, anyway.


No, I understand that getting consistent, repeatable, accurate (relative accuracy anyway) is what's important.

It looks like you've done the usual good job of covering all bases. Two big thumbs up!

I was just hoping to bring some discussion to some areas which didn't seem to be fully covered in the article (but perhaps I missed!).

Very very through as usual. BIG thumbs up!

With such a long time with no updates to the front page, I was starting to wonder if you guys were asleep. Silly me...

Love it, looking forward to more excellent articles!

Great article!

I think you'll need an engineer experienced in designing wind/water tunnels to do better.

I do think the grid would be helpful. A series of grids might work better. One deep grid creates boundary layers along the walls of the grid. This is basically air resistance. The deeper the grid, the thicker the boundary layers and impedence. I'm thinking several thinner grids would be more effective.

One other thing I'd like to point out, but may be an insiginficant factor is the pressure. Fans with more airflow will behave differently than fans with less airflow because of the pressure buildup in the case. The 120mm fans have to fight against more pressure to keep air from going out against it's airflow. To minimize differences in pressure, you could make the exhaust opening adjustable. Smaller for smaller fans, larger for larger fans. This might balance some differences between fan sizes. Of course, if we're just looking at 92mm vs. 92mm, etc, than there's probably little value in what I'm suggesting.

Excellent!

The hot wire is really the way to go. We use them in wind tunnel testing to make flow measurements for low-pressure flows without disturbing the flow too much. For high-pressure flows you can use a pitot tube, but a PC fan is nowhere near that point.

I'm sure you could experiment with a lot of different test configurations; I think the current configuration is a little bit overkill in terms of the volume of the case, but using conservation of mass to get all of the flow going in one direction is a good idea.

The most important advance I see here is the use of the hot wire. The 'turbine' anemometer you had before had a large hub which clearly alters the flow. For holding it up and measuring wind speed, it wouldn't be a big deal, but in this application you need something better.

I would suggest mounting the hot wire inside the test structure somewhere, as opposed to at the exit where the flow is expanding.

(IAAAE - I Am An Aeronautical Engineer)

See this post.

the last paragraph of page 6:

In the automotive industry, this is called a Mass Airflow Sensor (or meter). They have been used for quite a few years, and they replaced a similar device called a simple Airflow Meter. The airflow meter was a simple flap-door which was pulled open by the intake airflow. How much it opened was metered by a simple resistance device. The Mass Airflow Sensor is used for a similar purpose.....but is more sensitive to air temperature, and air pressure. As to which of the two devices would be a better guage of computer fan performance......hard to say. I second the motion that the sensor should be inside your fan enclosure. The car Mass Airflow Sensor is extremely sensitive......it is usually very close to the air filter, with a screen in front of the wire. Any change in airflow, such as removing the air filter, or leaving the air filter housing off completely, skews the readings. So much so in fact, the car engine may stop running, and the CEL (check eng light) may come on. But I do think this is an excellent way to compare fans.....oh. These devices fail frequently. I've got a few in the garage. Neat housing about 80/92mm in diameter, but I never found a use for them.

As to what causes the things to fail.......dust/dirt, the same thing that clogs up computer internals.

That was a lot of work, and some interesting results, thanks SPCR!
But testing the fans on the Ninja wasn't helpful - the Ninja is well known for it's low impedance. The 1 degree difference wasn't telling much.
To determine how these fans rank when coping with backpressure SPCR should have tested them on a heatsink with tightly packed fins (SI-128 maybe?).

Darn, now you have to completely redo all the previous fan roundups to account for this new method.

I'll give you till then end of the week.

........Aaaaaand GO!

I disagree and I might be biased, but I think Ninja is probably the most popular cooler here atm. And the test was useful to show that airflow becomes redundant at some point.

Why would we want to use heatsinks with tightly packed fins, when we can cool low impedence heatsinks with lower rpms?

The Ninja may be popular, but do I have to remind you the bad hit Rev.B takes on Socket 775? That's why I didn't get a Ninja, but a SI-128.
The performance level of a heatsink begins at the transfer surface between it's base plate and the CPU heatspreader.
From many posts on these forums it appears the Rev.B sucks at this matter. There are workarounds, I know.
I and other readers may want to know the results of SPCR fan tests on not-Ninja heatsinks.

stay away from these fans, bought one on friday and received nexus and noctua yesterday. Guess why! They should change the name to NoiseX Ixtrema. I used it on ninja as a replacement for stock fan and after xilence (very silent but poor airflow) and I was blown away but with the noise. To make it inaudible I've had to reduce speed to 40% in speedfan. I have Antec P182 and still experimenting with fans - have Antec TriCool, Akasa Amber, AC 12, nexus, noctua 1200rpm, xilence, scythe Flex E, scythe from Ninja (ADDA), SilentX Ixtrema.

You're still missing the point.

Once the airflow to any heatsink exceeds its heat transfer rate between CPU and fins, then any further increase in airflow is useless. The key point is that with the Ninja, which is truly optimized for low airflow, this occurs with 120mm fans spinning at 1000~1200 RPM --- and for most SPCR enthusiasts, even the quietest 120mm fans at this RPM are already too loud. Yes, you can get better cooling at 1500 or 2000 rpm with a Scythe Inifiny (for example) -- but would you want to accept that noise? The point is -- we don't care if you can get better cooling with higher flow or not -- it's simply too loud.

IMO, it's important not to overplay the "CFM is irrelevant" soundbite. Yes, after a certain point extra CFM has no extra cooling effect, but where is this point on the most common heatsinks? We don't know. Not everyone can afford a Ninja, and so heatsinks with higher-than-desirable impedance have to be used. As a result, it is very critical to know which fans can cope best with higher impedance, ie has a P-Q curve that intersects high on the y-axis, and still remain quiet.

I'm not missing your point, I completely agree that for the Ninja more than 1,200 rpm is a waste. But for heatsinks with more tightly packed fins, when tested with the same 1,000-1,200 rpm fans I would bet on a different ranking than on the Ninja. Which matters for not-Ninja owners, in order to buy the 'best' fan for their CPU heatsink. BTW, the 1,200 rpm at maximum possible load is not so bad if under light load the same 120 mm fan stays under 600 rpm, being temperature controlled from BIOS or other software.

It's too high is where it is.

I have to say that if you're looking for quiet CPU cooling, my recommended approach is still to get the best low airflow HS you can buy, and mate it to the best quality sound fan you can buy. Then make sure your case has very low airflow impedance and good airflow, and tweak the CPU fan speed by ear & an eye on the CPU temperature.

As for $, ok, high end HS range $40~60 and cheap ones can be <$10. Most people reading SPCR can afford to pay the difference. If they can't, they'll experiment with what they can get -- like we all do.

I don't think it's critical at all to know which "fans can cope best with higher impedance, ie has a P-Q curve that intersects high on the y-axis" -- all that is so easily compensated for by simple things like case airflow. You don't need to be an engineer to have a well cooled yet quiet system. Besides, fans of the same size (ie, 120x25mm) barely have any differences at all in pressure capability. (I know, I know, I will have to demonstrate that one day -- with a high impedance HS and a bucnh of fans.)

Good work on the article, MikeC.

I have to agree with the comments above regarding the test method, and that is it would be good to ensure an even distribution of airflow out of the exhaust aperture that is all traveling perpendicular to the aperture to get good measurements. I'm guessing that the 'baffle' in the middle of the case will cause some of the air to exit in somewhat of an upwards direction.

If you can sort out an exit grille/collimator to take care of the airflow distribution and direction problems, then you should be able to get back to only taking one airflow measurement instead of having to average many readings.

I would say that you don't actually need zero backpressure, just no more backpressure than a typical heatsink would provide. Hence maybe the straw idea could be tried again (perhaps using more/less, shorter and/or larger diameter straws to achieve the right backpressure). You could determine the appropriate backpressure by perhaps rigging up something to consistently (not necessarily accurately) the airflow _into_ a fan for a typical heatsink eg Zalman 7700, and then adjust your outlet backpressure of the test rig to match performance. Even matching fan speed may be sufficient.

There's probably a whole new article just in determining this!

But in the end I'm inclined to agree with you - buy a quiet, low-med airflow fan and make it as quiet as you can get away with thermally by undervolting it.

Cheers,
D.

I have a couple suggestions for reducing the turbulence effects:

1. Flip the fan. Use the fan as an EXHAUST instead of an intake. There's a lot of helical directional flow on the exhaust side of a fan, but very little on the intake side. By flipping the fan, airflow should be much smoother.

and

2. Move the test opening. Instead of placing the fan and opening on opposite sides of the air chamber, place them on the same side, seperated by a wide distance. This will eliminate any direct airflow path effects.

The Arctic Cooling 12L gave us very low airflow readings previously. Now the measured airflow is much more reasonable, although still lower than specified, and lower than the same speed but conventional design Nexus 120. This may be a real weakness of the Arctic Cooling open frame design; we suspect that some of the kinetic energy of the fan blades is dispersed and lost on the intake side before it has a chance to be directed into the output stream.
According to my experience Arctic fans have to be tested not from the "flow vs RPM" point of view but from the "flow vs noise" one. In this case they outperform almost any other fan out there (in open air) - spinning at higher revs but quietly.

Perhaps this is true if you only consider SPL but if you consider the overall sound signature, it's never quiet enough at any speed.

FWIW, the last few days have been a blur of fan airflow measurement experiments.

There's also been more feedback -- here in the forums as well as by direct email -- to this piece than almost any tech-geek article of this kind in several years. Usually these methodology pieces don't get read much; this one is already at >10,000 reads.

To address some of the many comments....

1) We did try reversing the fan. It may have helped but not much; we still had to resort to positional averaging.

2) In the test box, reversing the fan means the anemometer must go inside the box near the vent which is now an intake. The question is exactly where? The exact depth inside the box affects the flow readings tremendously. And what do we consder to be the cross-section? This is impossible to answer.

3) In response to some email suggestions, a long tube was also tried.

A ~2 meter tube, cross section about 130mm dia. End openings are 120mm. Both ends are formed from plastic food jars with their bottoms cut out.

Two flow-straightening grids about 1" depth with 1.5cm sq cells (like that shown in the photo I posted earlier) were used initially -- each inside the tube about 8" from the two ends.

Initial results --

There is still variance in LFM depending on position of hot wire. The edges generally give higher flow, the center lower. But it's not consistent -- some positions around the edge give >15% readings while others give <-15%.

Overall measured flow is lower than with published method, which still seems superior -- more consistent. Perhaps the flow-straightening grid has to be a LOT longer (deeper)...

The tube was made from 4 sheets of a long 2005 calendar, rolled up & taped. !!!

These pics will be amusing....









Later, the flow-straightening grid was combined and extended to ~4" depth and placed only on the fan side. Then the anemometer was positioned about 5" inside the other end through a hole drilled through the tube.



These changes made the flow a more even, but still not entirely consistent. The overall measured flow still slightly lower than our box.

So far, our published box + hotwire meter mutiple position averaged readings still appear the most reliable.

However, some very specialized honeycomb airflow straighteners are keeping the bazooka tube alive. It's possible the 1.5cm sq cells of my air straighteners are not good enough. This might do a better job: http://www.saxonpc.com/index.html I have some samples coming my way.

Respectfully, I think you may be missing a point as well...



The above is from your highlighted section with the higher speed fan test, which has a couple of important distinctions in it, which are later seemingly ignored in various conclusions... the fact is, there is some (perhaps significant) segment of the "SPCR enthusiasts", who are also enthusiasts first and foremost.

In other words:

1. they overclock.
2. they want a HSF (including fan) that can handle thier overclock
3. they want that HSF to be as quiet as possible, while handling #2.

...note that #2 does not necessarily jibe with SPCR's primary "silence before everything else" position, but the point I'm trying to make is that while your site provides a valuable service to the community, it tends to leave a large portion of folks who might otherwise benifit, hanging.

I don't come here to find out how to make an office PC silent. I come here to find out how to make my loud enthusiast/gaming PC less loud (as quiet as possible, at the temperatures I plan to generate).



...that may be the case, by why turn away readers who don't mind (are forced to live with) more noise in order to get higher temperatures under control? Wouldn't you want them to know the best possible solutions given thier configurations? Or do those folks just have to fend for themselves, because you really don't care?

I'm not ranting or flaming here, btw... just providing some food for thought.

Cheers.

I think Spanki does make a point, but there has to be limits. Taken to its logical end that reasoning means that SPCR should be testing every 100CFM fan, and every 60dBa HSF just because they are popular with enthusiasts. There are hundreds of "enthusiast" review sites out there which tackle that kind of hardware....SPCR should limit itself to only testing hardware that meets a certain minimum level of expected quietness to begin with. Time is the limiting factor in the number of reviews SPCR can produce. Even limiting ourselves to just testing "quiet" products we're still drowning in review samples.

This is straying from the topic at hand, but the way to handle the "CFM to cool" issue is a revision to the heatsink testing system. It would be pretty simple to generate a 'Max Useful CFM" number for each heatsink tested. just strap a high-flow fan to it, and slowly start cranking up the voltage until the temps stop dropping. It's one extra test to run, but it does produce a useful benchmark that can be used to compare one heatsink to another.

EDIT: Actually, if you just graph the results from our typical 12,9,7, and 5 volt testing results you can predict the "Max Useful CFM number" for any testes heatsink, with reasonable accuracy.

Rusty's response is right on the $.... but I want to add one more thing, re...


As already alluded in the article and in previous posts, CPU temperatures (and temps in general) are not just about a fan's output but the airflow to the components, and through the case. ie -- case airflow optimization. It is very possible to make very hot systems run cool enough and very quietly using the information already here at SPCR. But if you want to be on the cutting edge whether of OC'ing or silencing, there's no getting around "fending for yourself" -- you simply have to tweak, adjust and tweak some more to get the best results for yourself from your gear, with or without detailed analysis. The latter makes it easier, but you can't just drop in "recommended" parts and expect to have an ideal setup.

I agree - to an extent - with both replies above, so thanks for listening. I particularly like the idea of adding some sort of 'max usable CFM (or RPM)' rating for heatsink reviews, although I'm not sure that I agree that those numbers can necessarily be extrapolated from existing results from low speed fans.

But just to clarify my original comments, let's assume that what I really need (airflow-wise) for my particular heatsink, at my particular overclock, in my particular case is a 1500 or 2000 RPM 120mm fan (however that is determined)... so I want to find the quietest 1500 or 2000 RPM fan available.

My first thought might be "hmmm those guys over at SPCR care about noise - heck, they're the industry leader in such things, so let's go see what info I can find there"... <search>... <read various fan reviews>... <scratch head>... <look around some more>... "hmm... it looks like they don't bother even testing anything above 1200 RPM or so because that would be too loud for them - sigh".

I'm not suggesting that testing should be carried out to the extreme end of the spectrum (unless by some miracle or magic someone comes up with a 150-200CFM quiet fan), but there is certainly a large demand/desire for the 50 - 90 CFM range.

Even a "The best [Edit: "quietest"] high-performance 120mm fans we'd never be caught dead using" article would be useful. As a consumer, I don't mind a site/review/reviewer having or expressing some opinion/philosophy - but I also want to see the products that fall outside that range, so I can form my own opinions, based on that advice.

Just as an example, I recently read another review, which included some medium speed (and even some high speed) fans. Airflow, actual cpu cooling, measured dbA levels and subjective noise/sound comments were listed for all. I found this an extremely useful review, because it showed fans like the Sharkoon Silent Eagle 2000 that I hadn't considered or even heard of before. In that particular case, I wouldn't run the fan at full speed (too loud), but it out-performed fans like the Yate Loon and Silverstone FN121 when cranked down to the same dbA levels as those (which likely means that it was still spinning faster).