most CPU wattage and still be fanless - the Rusty challenge!

[dan]

Dear Rusty075,

"Nearly any Via C3 can be run fanless. Even the fastest ones, at 1.4Ghz produce less than 20watts of heat. With a decent heatsink (most Socket A clipped heatsinks will work) and even low passive airflow, fanless operation should be pretty easy to achieve."

what would be the most wattage you would run a cpu fanless? i ask b/c my current cpu is undervolted to run fanless at 14 watts. I'm thinking of overclocking it and i know power consumption will increase. how much can it increase before i need to put a fan on it?

[wumpus]

"Nearly any Via C3 can be run fanless. Even the fastest ones, at 1.4Ghz produce less than 20watts of heat. With a decent heatsink (most Socket A clipped heatsinks will work) and even low passive airflow, fanless operation should be pretty easy to achieve."

I have two Via C3 systems, and I will tell you right now, that is a tall order. You better be comfortable with CPU temps upwards of 90c for fanless operation on 14+ watts. That, or else truly MASSIVE heatsink assemblies-- not high end standard coolers like the SP-94, no sir, I am talking ginormous heatsink assemblies: on the order of Zalman passive case or fmah stuff.

There is a world of difference between low airflow and NO airflow (completely passive). Low airflow is child's play compared to passive. You wanted a challenge, you got it, just be careful what you ask for

[ChucuSCAD]

case and point..... thank you.... elvis has left the building..



Zalman Tnn500a


chucuSCAD

[Rusty075]

How much wattage you could run passively simply has too many variables to even be guessed at.

As wumpus points out, in a standard case situation, you're going to have to go really, really low wattage to be "passive"

And as Chucu points out, if you're willing/able to go to an extreme, a la the Tnn500, (or a Hush) you can go high wattage and still be "passive"


Passive cooling is the Holy Grail of Silence for some...but it's really not necessary. You can build a fairly conventional fanned system that is essentially silent.

My current rig is fairly conventional: XP-M 2500, a Seasonic PSU, an Ati 9500 VGA card, HDD, etc, etc in a standard case...and it's below 15dBa. (that's a rough guess, I don't yet have the real equipment to measure that number, but it is more quiet than a 7v'd L1A fan when compared from the same distance. At 7v the L1A should have a dBa in the 15-16 range) Essentially it's completely inaudible from the user position. So if silence is the goal, it can be done without the trouble of being fanless.

But...if you want to go fanless, for whatever reason, you have to manage the airflow very efficiently. Passive does not mean zero airflow. The heat itself will create the airflow, your job is to caress it into doing what you want it to do. Not easy, and there's no good guide for how to do it. Too many variables for each situation.


.....I did once have a completely fanless K6-2 550 system; CPU, PSU, everything fanless...and at 25 watts, that's a hotter, more power-hungry CPU than any of the C3's are....

And all I had to do to run it passively was submerge it all in an aquarium full of baby oil.


So ask yourself: Why is passive so important when silence is the real goal? and To what lengths am I willing to go to get it? once you have answers to those, then we can talk specifics.

[silvervarg]

Dan:
what would be the most wattage you would run a cpu fanless?

I run 20.4W passive with an off the shelf heatsink (and a few smart tricks).
With just convection air cooling and off the shelf heatsinks this is rather close to what you can do. With lots of fine trimming you might be able to reach 25W.
To get any higher than that you need to get a much better heatsink, something with heatpipes to the case, similar to Zalmans TNN-500 or HushPC or Fmah's homebuildt thing etc.

Rusty075:
Why is passive so important when silence is the real goal?

It all depends where you got your computer placed etc. IF nearly silent is not good enough then going passive is the next step.
It also comes down to the question: Why stop when you can go further?

[maxw]

case and point..... thank you.... elvis has left the building..






chucuSCAD

I had a via C3 667 cpu that was built onto a motherboard I and made a custom large heatsink to go on it and it may have run fanless but not under 70 degrees and the heatsink was far too hot to touch - I wasn't comfortable with that:)

[TheWesson]

I would put in a chimney that was as tall as practical. If you could just get 5 CFM going thru it you could cool a 100W CPU.

Sans friction the longer the chimney the faster the airflow - but in real life friction is going to put a limit on it pretty soon.

I'd use a Zalman 7000 and a 12cm dia meter-high insulated stack. Also a couple of 12cm case openings.

It would be absurd but, I think, doable.

By the way - I wonder - convection cooling arrangements like the Zalman TNN seem to go to no effort to make a "chimney" around the external heatsinks - does it just not work that well?

the wesson

[the_smell]

In theory, the universe being infinte, you can build an infinitely large heatsink and be able to coll what ever you want. Sorry got philosophical there!

I keep a nehemiah 1.2 GHz under 60 degrees under load with just a Zalman CNPS6000-CU and the airflow from the nexus 3500 PSU (which doesn't ever speed up ), not exactly passive but quiet enough to sleep a couple feet away.

[TheWesson]

If you want an (effectively) infinite heatsink, stick your computer in the surf. Pump-free water cooling.

Seriously, you could take the Zalman Reserator and glue the bottom of it to the CPU HS. I bet water convection would take away plenty of heat, and the Reserator has enough surface area to put the heat into the air ...

<thinky> Jeez that would just about work. Never mind the glue, the weight of the Reserator would keep it in place.

Or to be a little bit more sophisticated, use heatpipes - one end on the CPU and stick the other end into the reserator.

the wesson

[silvervarg]

TheWesson:
I would put in a chimney that was as tall as practical. If you could just get 5 CFM going thru it you could cool a 100W CPU.

What do you base this on? According to my calculations and my practical tests you can't get even close to this performance with a realistic setup.
Also the friction is not the biggest problem with a chimeny, it is to keep the air in the chimeny from cooling down on the way to the top of the chimney.

Seriously, you could take the Zalman Reserator and glue the bottom of it to the CPU HS.

Perhaps a slight violation of the heatsink maximum weight...

If you want to go to the troubles of building something with a special buildt case and heatpipes made to fit your motherboard and your special buildt case then a solution similar to Zalmans TNN-500 is a good option.
To get even better performance you can let the heatpipes attach on a larger surface to the heatsink and also build something chimney-like on the heatsinks. Even without these tricks Zalman can take a lot more than 100W of CPU heat.

[TheWesson]

TheWesson:
I would put in a chimney that was as tall as practical. If you could just get 5 CFM going thru it you could cool a 100W CPU.

What do you base this on? According to my calculations and my practical tests you can't get even close to this performance with a realistic setup.
Also the friction is not the biggest problem with a chimeny, it is to keep the air in the chimeny from cooling down on the way to the top of the chimney.

Okay then, insulate a 5-ft chimney.

You're right that the airflow from convection is pretty pathetic - I tested 120 watts in a cardboard box and an 80mm exhaust at the top had barely enough airflow to keep aloft 1/3 thickness of Kleenex. But I don't know how many CFM that is. Somewhat less than 5 CFM in that case is my guesstimate.

5 CFM if used efficiently would barely suffice for a 100W CPU.

Here is the calculation:
C/W (thermal resistance) of an airflow = 1.685 / CFM
Thus, 0.337 for 5 CFM.

Supposing a great heatsink, 0.25 base C/W.

Total C/W = 0.587. This means 59C rise for a 100W CPU.

If your CPU can run at 80C, then you're in business.

OK, so maybe 7-10 CFM would actually be necessary. At 10 CFM, C/W of the airflow is .17 and HS C/W is 0.25, total C/W is 0.42.

Toasty 62C cpu - definitely possible!

the wesson

[Rusty075]

...Here is the calculation:
C/W (thermal resistance) of an airflow = 1.685 / CFM
Thus, 0.337 for 5 CFM.

Supposing a great heatsink, 0.25 base C/W.

Total C/W = 0.587. This means 59C rise for a 100W CPU.....

Where do you derive the "1.685/CFM" from?

Wherever it comes from, it's wrong.

°C/W isn't linearly proportional with CFM. Nor is the relationship a constant.

There is also no such thing as a heatsink having a "base" °C/W. All °C/W values are at a given airflow Change the airflow, the °C/W changes, but not linearly.

The chimney effect has been discussed here before, repeatedly. Stack effect

[TheWesson]

Sure a heatsink has a base thermal resistance. It's definable. If the air around the heatsink was always at ambient temp (the concept of a fan moving air at infinite speed, so there is zero temp gain in the air across the HS), it would still have a certain base thermal resistance.

The base thermal resistance is rapidly approached; 80 or 100 CFM is very close!

Now do you wish me to derive the thermal resistance of an airflow?

Here we go.

Assume airflow rate of X CFM with perfect mixing.

Assume air exits the heat source at a constant temp (equilibrium). This means that in one minute the heat source will have heated X cubic feet of air by (exhaust-intake) degrees.

Specific heat capacity of air is quite close to 1 Joules/Gram. That is, 1 Joule to raise 1 gram of air 1 degree C.

Joules = watts times seconds. 1 Joule = 1 Watt administered for one second.

A cubic foot of air weighs 1.25 oz, 35.6 grams.

We calculate the formula as if we were raising the temperature of one minute's worth of air (the volume of the CFM.)
RiseC * Grams = Joules
RiseC * (CFM * 35.6) = Joules
RiseC * (CFM * 35.6) = Watts * 60seconds
RiseC = (Watts * 60)/(CFM*35.6)
RiseC = 1.685 * Watts / CFM

This has the interesting effect of calculating the C/W (thermal resistance) of an airstream as well.

C/W = 1.685 / CFM

There you go.

Of course C/W is nonlinear - it is 1.685 * (CFM ^ -1) ... !

That gives a law of diminishing returns.

At 80 CFM, the "idealized" C/W resistance from airflow = 0.02 -- trivial compared to heatsink/fan ratings of ~0.25 (a very good HSF).

Therefore we can estimate the inherent C/W of that heatsink (in air) at ~0.23. That would be the value asympotically approached as airflow approaches infinity.

Now this is obviously an estimate because we don't know whether an evenly mixed 80 CFM is really making it through the heatsink ... let's say it's only worth 40 CFM of evenly mixed air ... then inherent C/W of the heatsink is ~0.21 ... not a huge difference ... enough for back-of-envelope calculations, anyhow.

the wesson

[TheWesson]

We would probably want to regard the idealized C/W of X CFM as a lower threshold for C/W ... in reality, air won't be perfectly mixed.

In practical experience, it seems pretty well in the ballpark for case temps vs fan throughput, for example. That may be because in well designed high CFM case environments, there must be so much turbulence that the air is thoroughly mixed.

Besides mixing, what other adjustments would be needed? I suppose the idealized formula regards the HS as a point source ...or maybe the theoretical perfect mixing already takes care of the heatsink shape issue.

the wesson

[edit:]
PS Thanks for the linky, Rusty. Most illuminating. According to the numbers for the stack effect calculator it is not even slightly difficult to get 10 CFM in the stack.

[TheWesson]

I hope I'm not being too annoying here, but let's work an example.

Take the Zalman article, about countercurrent airflow, and let's see if we get some plausible CFM numbers.

Using the numbers for normal voltage for A64, we see a heat rise, under load, of 35C over ambient for this HSF. Let's say the A64 is 80W under load but 10W escapes via the socket and otherwise, leaving us with 70W to escape through the heatsink.

The C/W of the HSF system is 35C/70W = 0.5 C/W

Now, I happen to know that the Zalman 7000Cu is about as efficient as the best heatsinks, given the right kind of ventilation. So let's call it 0.25 C/W inherent to the heatsink (You would get 0.25 C/W maxing out ventilation on it.)

Thus, we must ascribe 5.0-0.25 = 0.25 C/W resistance due to airflow.

0.25 C/W = 1.685/CFM
CFM = 6.74

That sounds about right for an L1A downvolted into silence ...

Anyhow, I should shut up now.

the wesson

[Rusty075]

Sorry for the delay in responding, I was out of town for the holiday weekend.

Sure a heatsink has a base thermal resistance. It's definable. If the air around the heatsink was always at ambient temp (the concept of a fan moving air at infinite speed, so there is zero temp gain in the air across the HS), it would still have a certain base thermal resistance.

Emphasis mine.

You just answered my point. At infinite airspeed, the Delta T of the Air, and the Delta T of the Heatsink, both equal zero.

At infinite air speed (CFM), Delta T=0°.

0°C/W=0 (no matter what the wattage is) Therefore, the "base" thermal resistance of any heatsink is 0, if enough air can be applied. (So I suppose that technically you are correct, all HS's do have a "base thermal resistance, and conveniently enough, they are all zero. )

Here's an example: Short Media's review of the AMK PC76-1708.
In it they show a 0°C change between idle and load temps using a 68watt CPU with a stock heatsink How you ask? Airflow. Lots of airflow. 1708CFM through the case, in fact. But according to the "base thermal resistance" theory, he should be getting at least a 17° rise (and that would be with your example of the "base thermal resistance" of 0.25°C/W of the excellent 7000a)



That's the whole reason why CFM must always be listed when referencing °C/W...without the context, it's useless.


As a practical aside...show me a single technical document that supports the concept of a heatsink having a "base thermal resistance": White Paper, tech spec, photocopied page from an engineering textbook, heck, even a manufacturer's ad would be fine.


Your "base thermal resistance" and your "1.685" number are clearly related: without each other they are useless. Unfortunately both already are.


The single fatal flaw in your "1.685" assumptions is that the transfer between the heatsink and the air will remain constant as the CFM changes. It doesn't. That's the entire basis of heatsink shape design, and the reason while some heatsinks perform better at low CFM while other's do not, even though they may be comparable at higher CFM numbers.


Lets try some examples where there are fewer unknowns:

Socket-A Heatsink Roundup

This is a more useful test for several reasons: The CFM is more precisely known, Wattage is a given (unlike the A64's Thermal Design Power), and multiple tests are done with the same CPU/mobo/fan, thus removing the many variables there.

We'll use the data there to solve backwards for the mythical "base thermal resistance", using the supposed "1.685". Since the "base thermal resistance" is a constant for each heatsink design, the numbers should be same across the CFM range.

Repeating the calculations for each of the 4 tested heatsinks:

Using your formulae:

Whereas: °C/W of the HSF system equals the "base thermal resistance" plus the "resistance due to airflow"; and "resistance due to airflow" equals 1.685/CFM

You get that: "base thermal resistance" equals °C/W of the system minus the quantity of 1.685 divided by the CFM


To simplify, we'll call the "base thermal resistance": X, the calculated °C/W: R

X=R-(1.685/CFM)

From the review we'll use four different heatsinks, with four different CFM and their R values.


SLK800:
With 53CFM Sanyo, °C/W=0.23 X=0.20
With 24CFM L1A, °C/W=0.29 X=0.22
With 14CFM L1A, °C/W=0.40 X=0.28
With 10CFM L1A, °C/W=0.74 X=0.57

X variance of 285%


AX7:
With 53CFM Sanyo, °C/W=0.34 X=.31
With 24CFM L1A, °C/W=0.40 X=.33
With 14CFM L1A, °C/W=0.49 X=.37
With 10CFM L1A, °C/W=0.63 X=.46

X variance of 148%


The ALX results alone disprove your claims. If the resistance due to airflow is a constant linear variable, and the "base resistance" is a constant, how does the SLK800 outperform the ALX at high airflow, but is beaten by it by 7° at low airflow?


6000Cu:
With 53CFM Sanyo, °C/W=0.38 X=.35
With 24CFM L1A, °C/W=0.52 X=.45
With 14CFM L1A, °C/W=0.63 X=.51
With 10CFM L1A, °C/W=0.88 X=.71

X variance of 200%


MC462a:
With 53CFM Sanyo, °C/W=0.29 X=.26
With 24CFM L1A, °C/W=0.40 X=.33
With 14CFM L1A, °C/W=0.49 X=.37
With 10CFM L1A, °C/W=0.74 X=.57

X variance of 220%

Wow, an average variance of better than 200% So much for it being a constant.


Wesson, don't take this little lesson as a personal attack. Clearly you're a very bright guy; your heat capacity of air calculations were spot on, just misplaced. One of the best purposes of a place like SPCR is to dispel myths. Now granted, your myths are new ones, nothing of the scale of "aluminum cools better than copper" or some such, but its best to nip it in the bud.

Of course, if I've done something wrong in my logic or math, or if you can produce some technical support of either of your theories, please do enlighten us. (I did some googling to try to find evidence to support your suppositions before going through all the trouble to type this up. After all, if it were true it would save me all sorts of HSF testing. The only thing I found on topic in support was other posts, in other forums, by you. Sorry)

[silvervarg]

Another hint that Wessons calculations where wrong is that fact that he could easilly cool a 80W CPU with convection and a normal heatsink to acceptable temps. Since noone manages to get even close to this in reallity it had to be really wrong somewhere.

The lessoned learned from this is nothing new:
Check that what you have come up with seems realistic. If it seems unrealistic or highly questionable try to proove/disproove your theory.

In this case I think Rusty075 found the problem(s). Still I think Wessons approach from a theoretical standpoint was interesting and led the discussion further on the way.
He did mention a few interesting things that we could keep working on.

E.g. What effect will it have if I put isolation materials on my chimney?
At the moment it is made from thin transparent plastic material, but 99% of the time the box has the lid on and no windows, so it would not be a big sacrifice to isolate the chimney.
If it would keep my CPU temp to acceptable levels when summer heat arrives (room temp could become 35*C top) then it would defenatly be worth doing.

So far my 20.4W seems to be the highest anyone has cooled with convection cooling. Is there really onone else who has given this a good try?

[Rusty075]

Wrapping the chimney in dampening material shouldn't have any negative effects, at least as long as you keep the extra material on the outside of the chimney. Putting it on the inside would increase the air friction, and decrease the cross sectional area of the chimney, both of which would be bad things.

But I do question how much good it will do... what's the purpose of the isolation material? Isn't your setup basically silent as-is?

For a pure air-cooled chimney, I can't think of any off the top of my head that were more than what you're doing. Joesgarage11, in the thread I linked to, was using a water-and-chimney setup to cool a 60W+ CPU. But you might consider using water cheating.

[TheWesson]

Rusty, thanks for the compliment. I will have to disagree that with infinite airflow (surroundings maintained at a constant temperature) the thermal resistance of the heatsink will go to zero. The material of the heatsink itself, most importantly where it contacts the core, has a thermal resistance, and this is why HS designers use copper instead of aluminum (for example.)

The 1.685 number is valid w/o a "base resistance" in the case of a case. It predicts temp rise quite well.

You can graph the observed C/W of a heatsink as the CFM is increased and clearly see that it is declining asymptotically not to zero but instead to some other figure. I suppose there is some good thought experiment to demonstrate this too.

I will have to examine the HS results more closely, but I will admit immediately that the air in a heatsink doesn't have ideal mixing, there is a lot of resistance, all the heatsink is not at the same temperature, not all the air in the heatsink is the same temp, concurrent is less effective than counter current, and so on.

The 1.685 constant correlates very well with data points for case cooling, where the air is probably mixed a lot better, and resistance is less important (the air isn't being pushed thru thinly spaced fins.)

So it's a rough idea only for what air does in a HS. But it's an important concept and the abstraction is valid I believe.

the wesson

[Dirge]

I am also interested in running a computer with no fans at all. Initially I had thought the VIA EDEN Mini-ITX boards would be ideal. But have discovered they do get hot to the touch and rely on case fans for their cooling. To me this sort of defeats the purpose of a Fanless CPU.

Undervolted fans just don't cut it for me I need a truly fanless system.

How can this be done withought the use of exotic or expensive materials?

[TheWesson]

Rusty, the data points you gave offer a pretty good fit if you assume a loss factor of a certain portion of the airflow for each sink. That loss factor is partly air resistance and partly other factors I am sure. I tried to fit a loss factor by hand and this is what I got.

CFM-actual = rated-CFM / LossFactor

CPU HSF, Loss Factor
SLK800: 2.2
AX7: 1
6000Cu: 1.3
MC462a: 2

So this implies loss of CFM is high for SLK800, MC462a, and low for AX7 and 6000cu.

Well, I don't know what to think about the MC462a pin config - but the SLK800 does have close fins and the AX7 and 6000Cu have widely spaced fins.

Here are the rated "base C/W" of the HS's at 54,24,14,10 ... using the above loss factors.

SLK 800
0.160056604
0.135541667
0.135214286
0.1993

AX7
0.278207547
0.259791667
0.249642857
0.2915

6000cu
0.308669811
0.358729167
0.353535714
0.49095

MC462a
0.226415094
0.259583333
0.249285714
0.403

Obviously 54 CFM and 10 CFM are not a great fit. Probably air resistance is not linear with velocity but instead with square of velocity.

Here are the graphs of the C/W vs CFM that you provided. At least for the AX7, it should be clear that it is not asymptotic to zero. For the SLK800, consider the correction factor due to air resistance and it is plain we are seeing the elbow of the graph not the trailing end.
C/W vs CFM

The following article and graph clearly show a heatsink bottoming out at about 0.17:
http://www.overclockers.com/tips358/

the wesson

[TheWesson]

Another hint that Wessons calculations where wrong is that fact that he could easilly cool a 80W CPU with convection and a normal heatsink to acceptable temps. Since noone manages to get even close to this in reallity it had to be really wrong somewhere.

I didn't say a normal heatsink. I just said convection. On this very site, in the thread quoted above, somebody used a 4' convection tower (with the heat pumped it via a waterpump) and no fannage whatsover. Big heatsink in the tower.

You could probably do the same with a Zalman 7000, if you could get all the CFM from tower convection actually going thru the heatsink.

This would imply the mobo sideways inside the chimney.

Zalman TNN is entirely fanless. Will accept 80W CPUs. Mostly cooled by passive convection; a bit of radiation.

the wesson

[Rusty075]

Wesson, I do see the point in what you're trying to explain, but you're stuck too far in the theory, and ignoring all the evidence in an attempt to save it.

The empirical evidence trounces both segments of the "thewesson theory of heatsink performance" There is no base thermal resistance, and °C/W cannot be computed from CFM numerically.


There is some value in detailed numerical analysis of heatsink performance. Manufacturers do it during development. (for some excellent examples of enthusiast number crunching, check out Procooling.com's waterblock testing. Outstanding work.)

The issue at odds here is whether you can use some out-of-thin-air (pun!) number to estimate the performance of a heatsink, which is what you proposed. Clearly that's entirely incorrect. There are other variables at play. The simple truth is that you cannot estimate the performance of a heatsink without conducting tests of it. And once you've tested it, the value of your guess-work drops to nearly zero.

And until proven otherwise, the empirical evidence for heatsinks having no "base thermal resistance" trumps the baseless theoretical musings. A more rigid analysis of the °C/W/CFM curves shows them all dropping asymptotically to zero resistance at infinite airflow, given the margins of error involved.


This may be going off-track, so to recap:

"base thermal resistance" = wrong
"1.685/CFM" = correct, maybe, but not for heatsinks.

Now lets just move on. Please

[Dirge]

Dear Rusty075,

"Nearly any Via C3 can be run fanless. Even the fastest ones, at 1.4Ghz produce less than 20watts of heat. With a decent heatsink (most Socket A clipped heatsinks will work) and even low passive airflow, fanless operation should be pretty easy to achieve."

what would be the most wattage you would run a cpu fanless? i ask b/c my current cpu is undervolted to run fanless at 14 watts. I'm thinking of overclocking it and i know power consumption will increase. how much can it increase before i need to put a fan on it?

Maybe everyone could take another look at the original post and offer up some suggestions

[AZBrandon]

Maybe everyone could take another look at the original post and offer up some suggestions

Back on topic, since I have an unlocked processor (mobile barton) and can control the case and HSF with SpeedFan and even shut them off with it, I did some testing tonight. Unfortunately, since the mobile Barton isn't officially supported by my motherboard, it runs it at 1.55v, so that's as low as I can go.

Anyway, even at 1.55v, I was able to run 5x166 (830Mhz) with just the stock heatsink and no active cooling except the power supply fan running. CPU temps leveled off around 57C with prime95 running, which is about as hot as I'm comfortable with. I'm positive that if I had full voltage control I could have likely gotten the same results at more like 1.0 - 1.2ghz somewhere around 1.15 - 1.2v. The mobile barton seems to have quite a lot of processing power and low heat potential.

[TheWesson]

The thermal resistance of a simple heatsink, a small block.
Calculating the C/W of an area of material

A block of copper .0001 meter square (1 sq cm) 0.01 meter (1 cm) thick - a small block of copper heated by the CPU - shall have a thermal resistance of 0.01 / (390 * 0.0001) = 0.256 C/W. That is the resistance through the block, of course, from one end to another.

OK, no more from me on this subject.

the wesson

[silvervarg]

Dirge:
Undervolted fans just don't cut it for me I need a truly fanless system.

How can this be done withought the use of exotic or expensive materials?

Check out what I did in this thread.


I already do have a passive setup, but we are aproaching the first summer for this system, so something that can improve cooling would be nice.
In my current setup I have a smart fan controller with temp control, so if temps climbs higher than I like a fan turns on and spins slowly. This ensures that the system will not crash or experience any problems.

If air in a chimney is cooled down a lot on its way to the top of the chimney the airspeed will drop considerably. So the idea with insulation on the outside of my chimney (still inside the computer) is to increase the CFM a bit with no added noise.

Do I want to insulate the part closest to the CPU or should I skip that to avoid having very hot metal close to the insulation material?

Can anyone (TheWesson perhaps) do any calculations on what it would do to my CFM if I add insulation?
I have no way of measuring my current CFM, but a guess on 4-5 CFM is probably close. An answer with a magnitude answer would be close enought.
E.g. Will I get 100%, 10%, 1% or 0.1% more CFM with insulation?
My chimney is 30+ cm tall with exhaust hole as a 92mm fan and intake hole is a 80mm fan (with a stopped fan blocking part of the intake hole).

[Rusty075]

Silvervarg, figuring out how many CFM insulation would add, if any, would require lots more information: Temp gradiant from inside the chimney to the outside, R-value of the insulation, area of the insulated portion, a very precise measurement of the current CFM, etc, etc. It would almost certainly be easier just to try it and see what effects it has on your temps. If I really, really had to guess, I'd bet that the CFM increase will be pretty small, probably small enough to only have a couple of degree impact on CPU temp. And as both the ambient and the CPU temp change, the gradient, and therefore the CFM will adjust dynamically, making any calculations even more useless.

Just try it...you'll be adding new knowledge to the community as a whole.

wesson, theory is still trumped by reality: The SLK948 and the Zalman 7000a both test out at °C/W's less than your theoretical minimum. As do many aluminum heatsinks. Using your math the lowest an Al HS should be able to get is 0.42°C/W. (The trouble is the ΔT, is the airflow is high enough, the gradiant stays at zero. With enough airflow you could cool the CPU with no heatsink at all. )


To get back on-track, as Dirge suggested, I think my first response in this thread is still the most basic answer:

You can go as high a wattage as you want with a passively cooled system, but the lengths you'll have to go to get progressively more complex. For most situations, the results you can achieve with effectively-silent fanned cooling are easier and more effective.

[Dirge]

A truly fanless system would be the holy grail of computer silencing and none to easy to achieve.

What I would like are some recommendations on how to do this. I know its a major challenge but the rewards are great.


I don’t mind underclocking/undervolting or even buying a VIA C3 in order to do this. My only problem is that i dont have access to some of the more interesting heat sinks here in New Zealand.

silvervarg has a great looking system but I am unable to buy the massive cpu heatsink he uses.

Can someone suggest a fanless rig with off the shelf parts?

[fmah]

More realistically, the temperature difference is proportional to the inverse of the square root of the velocity.

So more like DT ~ 1/ sqrt(v)

Various heat transfer relations show this, also the power might not be 1/2 (i.e. square root of v) but some other number less than 1. That shows that as you continue to increase the velocity, the change in DT is less and less, which is what you would expect.

As far as fanless operation, well for heatsinks in the size range of a CPU, the natural convection thermal resistance for a 3" long piece seems to range around 2-5 °C/W. That would mean if your processor is 20W that would likely overheat if you are closer to a 5°C/W heatsink. Of course the thermal resistance will vary and could be a very large number, it all depends on the heatsink shape/construction.

A good thing to try, I personally think, would be a AMD mobile chip and undervolt and underclock it. You can get good performance for this kind of chip. I think someone has tried to mount a Zalman heatpipe VGA cooler to a CPU, if you can manage this somehow that would be pretty good.