I correction/clarification... I just went back and checked your 120mm fan roundup and you do in fact include a couple 50+cfm fans (and/or higher than 1200 RPM), so I wasn't trying to imply that you didn't - I was just pulling some numbers out of the air for my example above.

Unfortunately, this isn't quite how heat transfer works.

I'll try to explain this briefly and concisely, but I may fail. Please bear with me. I have my heat transfer book right here, so I'm not pulling this out of my ass.

For 1D heat transfer through a simple solid, the overall heat transfer rate depends on the temperature difference between the hot and cold ends, the distance in the direction being considered, the cross sectional area, and the thermal conductivity of the material:



It is often popular to redefine this equation as a thermal resistance:



The nice thing about the 'resistor model' is that when you move on to a more complex system (say, a computer heatsink) you can sum the resistances, and rearrange the above equation to get the overall heat transfer:



Lets look at a (simplified) computer heat sink. There is a thermal contact resistance at the CPU-HSF interface, a conduction resistance in the heat sink itself, and a convection resistance on the fins:



The contact resistance is a bit difficult to determine. For an aluminum heat spreader interfacing with an aluminum HSF using Dow Corning 340 grease, the textbooks lists 0.07 m^2*K/W as a value.



The conductive resistance is determined by the material(s) used in the HSF, and the geometry of the base. Lets say that the material is aluminum, k = 238 W/m*k:



Now comes the hard part - the convection resistance at the fins. This takes two parts to build up.



The area of the fins is easy enough to find. nf, the fin efficiency, encapsulates that improvement that the fins give in terms of heat transfer over a simple flat surface. It's going to be something greater then one, and with about 4 more equations I could show you to calculate it.

The big part is h, the convection coefficient. It is determined based on the characteristics of the flow past the fins. Lets assumed the flow is turbulent, since the fins have square edges rather then rounded ones. For turbulent flow past a flat place



What does this all mean?

The Reynolds number, Re, increases with velocity, and increasing Re increases the convection coefficient, h. Increasing the convection coefficient decreases the thermal resistance of the convection portion of the heat transfer, Rconvection. Decreasing Rconvetions will decrease the total thermal resistance, thereby increasing the overall heat transfer rate between the CPU and the air.

Now, you will reach a point of diminishing returns where Rconvection << (Rcontact + Rconduction). At this point, increasing the flow rate over the fins will begin to have less and less of an effect, but to make the simplification that it is "exceeding the heat transfer rate between the CPU and the fins" ignores the fact that indeed, the heat transfer rate between the CPU and the air does depend on the sum of all of the thermal resistances inbetween. It's all one connected system.

Reference: Ineropera, Frank P. and DeWitt, David P. Fundamentals of Heat and Mass Transfer, 5th Edition. John Wiley & Sons, 2002.

Thanks for that - I'd been feeling deprived by no new SPCR article for so long, but clearly you've been busy.

Here's another view on the "no improvement with faster fans beyond a certain point" phenominon. The short version is that it is diminishing returns - if the heatsink fins are only a little warmer than the case air, you don't gain much by blowing more air.

In much more detail: you can make a strong analogy between heat disipation and electric circuits with resistors. Voltage becomes temperature, current becomes heat power, and resistance becomes thermal resistance. Then, roughly speaking, the heat flow from your CPU is like a current flowing through a series of resistors:
* between chip and outside of chip package (beyond our direct control)
* between chip package and base of heatsink (thermal grease)
* between base of heatsink and fins of heatsink (CPU heat sink)
* between fins and air in the case (CPU fan, and/or general airflow in the case)
* between air in the case and outside air (case fan)

At each step, there is a drop in temperature. This temperature difference is equal to the thermal resistance times the heat output of the CPU. Generally speaking, the steps where there is a large temperature drop (high thermal resistance) are where you'll get best return for your efforts in improving your CPU temperature - e.g. a 10% improvement in a large thermal resistance can be more useful than a 90% improvement in a small resistance.

The CPU heatsink fan is operating at the fourth step (heatsink to case air.) Doubling the air speed will (perhaps) halve the resistance of this step - but if this resistance is already small compared to the others, you won't notice any improvement for your extra noise - it is the difference between a CPU running 30 degrees above room temperature and one running 29.5 degrees above room temp. If, for example, your case is poorly ventilated (large resistance between case air and room air), you could get a much larger improvement for the same noise increase by speeding up the case fan instead.

(I'm trained in physics, not thermal or electronics engineering, so there are probably practical complications I haven't accounted for - most significantly, my analogy doesn't account for the directionality of air flow.)

P.S. When I saw your 'mystery photo' at the start of the article, I thought its purpose was to impede airflow.

I knew what it was for right away, but then I saw an episode of Mythbusters where they used bundles of plastic straws for essentially the same purpose.

Jumper & Filias Cupio:

I bow to my accusers and admit my guilt -- I did knowingly oversimplify a technical process. I promise not to simplify so sinfully again.... at least until my next article.

However, I stand by the practical implication of my simplification: For most PC thermal issues, beyond a certain airflow (the amount of which varies from case to case with many factors not easily controlled outside lab conditions), a point of greatly diminished cooling gain prevails, while the price paid in increased fan noise continues to rise. (Actually, for most silencers, this really means any "extra" airflow is worse than useless.)

The absence of any details about the tools used or pics / clear description of the setup makes it impossible to make a fair assessment of the roundup you linked to. The reviewer tells us, "After gathering the testing equipment, I formed a methodology that would provide extremely consistent and telling results".... but he doesn't show us any real evidence that this is true.

I suspect that he's just as prone to human error as the next guy, but aside from that, I have no reason to believe he just made up all those figures either. I'm not particularly interested in whether (for example) his dbA numbers or CFM numbers are exactly correct or not, as long as each fan was put through the same tests, it should allow you to compare them relative to each other.

Maybe you guys could add some of those fans to your own testing to veify or disprove his claims? That would be great. I noticed that you have the ebmPapst 4412FGL and 4412FGLL in your 120mm roundup, but not the 4412FGML, which shows very similar characteristics to the Sharkoon mentioned earlier in that review.

About those higher CFM fans... I would too like to see that you test those. Because... I think that many of them can perform really well when slowed down. For example I have an two ADDAs fans which are "loud" on 12v but perform well (quiet and moves still lot air) on 5-7 volts. And because they are temp controlled they normally run about 5 volt but after looong gaming session they probably run about 9 volt. And I can stand that because my silent but overclocked rig otherwise would crash. And for gaming it's not so big deal to have a little noise.

I'm pretty much not an engineer at all, but one thing comes to mind;

Why are you testing the airflow on the push side of the fan? The same amount of air has to go into the fan, and on that side you don't have to deal with turbulence.

Just an idea, I suppose.

[added later]

Whoops, I'm an ass; I missed the first page there. Go about your business.

Exactly like MikeC has said......the vast majority of 120mm fans blow the same amount of air at any given rpm. I've suspected this for a long time, and the article just re-enforces this belief. So what SPCR has to concern itself with is just how quiet each fan is relative to one another, at different rpm levels. The CFM ratings really don't matter. If you want more airflow go for the fan that turns faster.......don't believe the SilenX-type crap that their 120mm fans push more air at lower rpms. Just doesn't happen.

Excuse me but if you want silence and high cfm (for OCers) isn't the solution to use bigger fans (25 cm)?
I mean like with the Sharkoon Revenge case (25 cm + 14 cm fan)
http://www.sharkoon.com/all_eng.htm.

Ok you can't put a 25 cm fan on the CPU HS but still bigger fans like 14cm with the new IFX-14 from thermalright or watercooling is probably a better way to go. Maybee HS for 2x 12cm fans stick side to side is better also.

Hi, I'm new to Silent PC review Forums (say that 10 times).

Interesting article and discussion.

I am an electrical engineer, but work in the industrial control field. In our applications we always have the need to measure process exhaust flow, either with permanently installed instrumentation (venturi, s-tube), or spot-checks with handheld s-tube or pitot-tube.

The rule of thumb is to take your measurement in a straight section of duct, preferably round duct. You need a straight section of duct at least 10 duct-diameters long before, and 5 after, your point of measurement. This will straighten out the flow and make the reading more accurate over a wider range of flows.

The velocity (and thus flow) is zero at the wall and maximum at the centre of the duct, so you need to find the average velocity by taking 20 measurements in two traverses across the duct. Omega has some basic, but practical information on various flow measurement methods:

http://www.omega.com/literature/transac ... g.11_l.GIF

For a round duct, assuming that the flow is symmetric, the measurement simplifies to 5 readings taken between the wall and the centre of the duct.

The full article:

http://www.omega.com/literature/transac ... -DIFF.html

(The article is aimed at differential pressure methods, but the basic practices still apply to other methods.)

The cardboard pipe seems to be the correct way of doing the flow measurements. Unfortunately this will not eliminate the need for multiple measurements. Some sort of jig with preset insertion depths should make the job pretty easy though.

If the flows measured are within the same range, you might also find that the average flow is always measured at one particular insertion depth. This could easily be verified by testing the lowest and highest flow fans at the five different insertion depths, and finding the insertion depth where the average flow is measured for each of the two fans. If the 2 insertion points turn out to be one and the same, the probe could be secured at that insertion depth for all future measuments in that range of flow.

I was bored. I figured you knew what was really going on

Maybe SCPR could use a '(more) technical (then usual) library' section with articles on heat transfer and such?

klipkop --

Thank you very much for the links and for your helpful, knowledgeable comments. The info you provided is invaluable for SPCR -- I've already copied & pasted your post to my reference fan folder.

+1. Looks like your tube could easily be modified to provide the exact test setup that klipkop has described as industry standard.

This kid of discussion is why I love reading SPCR!

Good work, SPCR!

Quick edit needed at the bottom of the first page: "We'd have find our own way."

I don't care about fan voltage. I don't care about cfm. What I do care about is noise and temps. What I'd like to see is comparison of fan curves on a temperature/noise plane. Hook them up to a ninja and see which one farest best at any noise level, ignore voltage and cfm.

temps are strongly correlated with cfm, so actually you do care.

I also think we need some sort of backpressure (this topic got me since "fire and ice" ). Since free air cfm is the same for all makes, what about the ability to cope with backpressure? Here's my list of most common fan positions and the obstacle in their way.

1. case exhaust: grill
2. case input: none
3. HSF push: heatsink
4. HSF pull: not widely used
5. PSU: not the scope and 90° redirection

I don’t think it’s an issue of ‘not caring’ at all. I’ll even say that the issue of ‘caring’ is completely moot in this discussion. There are plenty, too many, perhaps, of sites devoted entirely to the issue of overclocking and all that comes with it. This particular site and its editor have one priority above all, and that is the issue of maximum suppression of noise. It would be a sad day shall I ever log in here to find that Mike and the gang have strayed the course and lost focus over what has made this site thrive.

Having said that, I must note that even hard core overclocking enthusiasts can and do reap benefit from the work that is performed here. Hence there really is no need to adapt the testing methodology nor the point of view from which the editors write to please said enthusiasts. But even if some adaptation to please others is considered, my opinion is that nothing should change if said change would be at the expense of the primary objective.

I have read and lurked around countless sites and forums over the years, yet this is one of only a handful (maybe 2-3) that I faithfully return to and read every single article and review that is put out. The sole reason is the focus and passion which is placed on the value of silence. Everything else is secondary, and should always remain so.

Take care,

I'm not asking for a change of direction to cater to the overclocking crowd, I'm just looking for the best information on noise possible - so 'secondary' doesn't (shouldn't) be equated to 'ignored', 'moot', 'irrelevent' or 'nonexistent'.

The only way to achieve complete silence is to have no moving parts - that's easy - find yourself a passive PSU an then throw out all your fans, CD/DVD drives, floppy drives and harddrives.

Obviously this is not always practical, so 'some' level of tolerance for fan noise must be factored in, to achieve 'some' level of performance.

That 'some' level of hardware performance ranges from being able to boot some basically functional computer to run the desired applications at some minimal-level of application performance on the low end to the gamer/overclocker/enthusiast-level performance on the other end, with all other levels in-between.

That 'some' level fan (and other component) noise has a coresponding and similar scaling to the above performance level.

I assume you don't come here to figure out how to build a completely silent computer with no fans or other moving parts (?), so I'm guessing that you have 'some' level of tolerance for compnent noise as well - based on the applications you intend to run and some minimal level of performance you expect to get from those applications as well as other things like the ambient noise level of your environment.

Unless you're (literally) deaf or otherwise just don't care about noise at all (in either case, this site would mean little to you), then we're all just trying to find that balance.

So we're just talking about a difference in degrees here (sorry for the pun). Should SPCR only concern itself with completely silent PCs with no moving parts? If not, then what's the cut-off above that level?

Whatever can be achieved below 20dBA?
What about 25dBA?
27?
30?

What if my ambient air temps are really high in the summer because I can't afford or don't own an air conditioner, but I'd still like my computer to boot and run my applications?

What's the range of audience that the performance of a completely silent PC will satisfy? Obviously this number gets larger as the performance level goes up and starts getting smaller as the noise level goes up accordingly.

As mentioned in my first post, this is all just food for thought.

It has been stated time and time again that SPCR considers an overall noise level of >30dBA @ 1m too noisy to be considered "quiet".



actually SPCR is one of the few places where there is a non-trivial amount of information about this, so yes that is part of the attraction.



um, increase your fan rpm/diameter?

Personally I have a big problem with helping people who run 150W+ GPUs in SLI/CF and then expect it to be silent without massive heatsinks or elaborate watercooling solutions; that just shows a basic ignorance of fundamental physics.

Ok, so the cut-off is 30dBA - thanks.



...just to clarify, I wasn't trying to suggest that that type of information wasn't available or desirable, only that the audience for this specific type of information is relatively miniscule as a percentage of the audience that otherwise benifits from this site.



I'll assume for the sake of discussion that you weren't implying that I am ignorant or wouldn't come up with the obvious "increase [my] fan rpm/diameter" answer to my hypothetical question on my own...

So let's assume for a moment that I had that flash of brilliance and after working on my case airflow as best as possible (reducing the number of wires and such hanging in the airstream, etc.) I go looking for the quietest fan possible that suits my cooling need... bah... I don't need to repeat myself, you can see my previous examples and comments in this thread if you're interested.

Just for the record, my system has a modest 2.66GHz cpu, and a single 7900GS vga card in a P182 case.

I'm thrilled to see that SPCR is taking steps to improve thier testing methods in order to provide the best information and reviews possible. They have hundreds of fans sent to them for testing so - as time permits - I'd love to see reivews on more of them. (<-- period).

Cheers.

jaganath has some good points, and I certainly appreciate einsteingodel's comments about why he remains a faithful SPCR reader.

Getting more to the nuts and bolts of Spanki's comments, here are a few general comments --

1) The fan roundups we've done already include MANY different fans; some mid-speed fans are represented. We will probably never test any high speed fans unless they are quiet; from experience, I can say they just don't undervolt well, the noise character never gets good (usually). From our point of view, we don't want to waste our time or that of our primary audience, who will not be interested in 120cfm fans @ 52 dBA @ 1m. BUt even among the fans we've tested there are models which far exceed 50 cfm -- and we have not only noise measurements but recordings and descriptions at 12, 9, 7 and 5V --- AND RPM and aiflow measurements at each of these voltages. You don't think this info gives you enough info to work with?

2) The gist of achieving a good balance of cooling and noise (whatever that balance is for you) usually involves more than one fan. The noise of fans is additive (tho not linearly so), so you want each fan to be significantly lower than the 30 dBA @1m we use as a marker. In some of my own systems, there are as many as 4-5 fans -- yet none are louder than ~ 23 dBA @ 1m. It IS a matter of juggling and balancing many factors.

3) I know that if I wanted to OC' as much as possible and use a super hot vidcard, I'd be prepared to spend more $$ on the best heatsinks and make sure the case airflow restrictions are as minimal as possible. Then I'd still go for the fans with the best acoustic signature -- never mind their ultimate speed or airflow -- and then experiment. I agree with jaganath -- if you want to OC a hot system quietly, there's no way to do it without better heatsinks. If your assessment of a particular HS is that with your CPU you'll need a fan that blows more air than the fan we tested the HS with, then you'd do well to look at the low speed fans we describe as having great noise signatures, then buy a fan in the same family, rathed for a higher speed. (The L and M Yate Loon 120mms come to mind.)

...well, 6 at least, plus some variants of those (doubling the number to 12 - in the 120mm roundup) .



I'll take your word on that (not undervolting well, noise character). As I mentioned (or at least implied), I'm not (personally) particularly interested in the "high-speed" fans either...



Yes - it was exactly the type of information I'm looking for - For the (relatively few) fans included in the review.

I did add a clarification in my post at the top of this page that you did have some 50+cfm fans (including the Antec Tri-cool, which was extremely useful to me, since I know what the Tri-cool sounds like, at different levels).



Yep - I agree with all the above, in particular the part about acoustic signature. For example, I happen to disagree with your (site's) particular appraisal of the Scythe fans... based on your recordings (and recordings at another site), I find the tonal/motor noise fairly irratating compared to some of the others that you gave worse marks to...

And that gets to the heart of my (only) real point in all of this - if your review had twice as many fans in it, I'd have twice as many fans to listen to and evaluate based on my own criteria.

Obviously, there are only so many hours in the day and the amount of testing you do on each fan takes a lot of time - I understand that, so consider my request for "more" as just the obvious extension of: "Great review! what about fan X?" Ok? .

What prompted me to even sign up to the site and comment was this quote from you:



...taken literally, comes off as elitist and might give some people the impression that your entire HSF section of the site could be replaced with one page that read:



...that would certainly save you a lot of work at least . (in case it's not apparent, the above was meant as humor - heck, the Nexus only finishes 3rd best in your roundup (though it might find it's way back to the top with the new measurements)).

Seriously though, I'm sure that wasn't your intent and I do appreciate your taking the time to respond and clarify your views and the mission of the site. Thanks.

It seems that the move from the turbine to the hot-wire anemometer turned out to be a trade-off. On one hand, you solved the rotational direction problem in the original setup and the impedance problem in the newer setup. However, now you have to take multiple measurements and average them.

The nice thing about the turbine method is that it doesn't care about cross-sectional variations in velocity--within its aperture. Everything that goes through it works to turn the turbine. It's a shame that nobody makes a turbine anemometer with a diameter of 120mm or greater. That would truly be the best solution.

I wonder if it would be easy enough to make one out of a 120mm fan. After all, what is a turbine anemometer if not a fan with a known CFM/RPM relationship and a way to measure its RPM? If there were such a fan with published and trusted CFM/RPM characteristics, one could probably modify it to (almost) freewheel and use a fairly simple circuit to measure the RPMs. Even if such a fan does not exist, a similar approach could be used to compare fans, producing results such as Fan A having 2% higher airflow than the Fan B at 800RPM.

While I appreciate your guidance that the CFM/RPM relationship is not nearly as important as the dBA/RPM relationship, it still has some merit. Often you conclude that the acoustics of two fans at the same speed are very similar. If further analysis reveals a 5% difference in airflow at this speed, then that result becomes important in achieving the goal of quiet cooling.

After all, the ideal fan metric for SPCR readers is CFM/dBA in conjunction with subjective acoustic analysis. One of the primary goals of the SPCR team should be to hone in on a method for measuring this relationship for any fan. I'm not convinced that this new apparatus is the correct approach, but it is better than the old method.

Given that there is such a strong correlation between RPM and CFM, I suggest that SPCR run a large number of fan 'microreviews': decide a set of RPMs (perhaps 2000, 1500, 1000, 700, 500) and for each fan record the sound level in open air at each of those RPMs which it is capable of running at. With a bit of coding, the human intervention could be reduced to swapping out fans and typing the fan name/model into the computer. The promising sounding fans can then be put aside for a more in-depth review.

P.S. Thanks for the detailed post about thermal resistances etc., Jumper. It wasn't there when I started my own much simplified post on the resistance analogy, else I wouldn't have bothered.

I think experiment #2 should be used in conjunction with the hot wire measurements since all cases (let alone heatsinks) cause flow restriction and knowing which fans cope better when faced with backpressure could be a real aid for choosing the appropriate fan for a particular situation.

And I'm not sure how having the fans tested as extractors is worse than inlet fans for measuring the airflow at the rear opening. Why does the anemometer need to be inside? Air still has to flow through the same opening.