(NVIDIA) Fan-Based-Heatsink Designs are probably wrong. (suck, don't blow ! heatfins direction)

Why not use compressed/expanded air for this purpose ? Using a piston compressor to compress the air to a few bars, the air gets quite hot, then let it go through a heat exchanger to get rid of most of the heat and cool the pressurized air closer to ambient temperature.

Let the air expand to normal ambient pressure and the air temperature is now well below ambient temperature and let it flow through semiconductor heatsinks to the environment.

To avoid problems with dust and condensation, a closed loop might make sense, but of course, now the heat exchanger would also have to dissipate the heat from the semiconductor. However, the heat exchanger can be remotely located and it can have much higher temperatures than the semiconductors, getting rid of the heat into the environment would be easier.

More to the point if the heatsink fins are not thick enough to conduct heat away from the thing being cooled it doesn't matter how easily you can push air through them. Equally it is no good if you get perfect laminar airflow since then only the air touching the surface warms up and the core air remains cool. So you have to have some turbulence and opposition to free flow but the tricky question is how much is enough?

Something like this might be close :

====o ====o ====o ====o ====o ====o ====o ====o

(slightly tighter together than ASCII art will allow) Airflow from left to right with a blob on the end to mix the air up.

But also very probably wrong. The volume of air going through the heat sink is proportional to the amount of cooling you get for a given design so there is a definite bias towards not blocking off half the free air flow. I would guess at something more like allowing 2/3 to 3/4 of free airflow as about the best depending on the exact heatsink geometry. It could easily be higher - easy enough to do the experiment.

I suspect the perfect shape for an optimum heatsink is rather more complex than the typical fins we get but the designs used at present are good enough and much easier to engineer. Heat pipes have helped enormously with the latest generation of quiet heatsinks.

It is a sobering thought that high performance CPUs often have a heat output per unit area that exceeds the tip of a soldering iron.

Regards, Martin Brown

You just invented the refrigerator.

Who would want to carry out repairs or mods on such a machine?

I know, I know.

The point is that by using compressed air, problems with hazardous substances and circulating liquids can be avoided.

The problem is how to solve the problem with humidity in the air, which might condensate or even form ice, when sub-zero air temperatures are used. The situation gets complex during startup, reboot, shutdown and especially during a power failure. If the humidity can be removed somewhere before expansion, the situation is greatly simplified.

I tried that on an old AMD and got the same results; the CPU was much hotter with the air being pulled through than it was with it being blown through. One of the electrical engineers at work explained it this way: If the air is being pulled through, most of the air is moving through the fin area closest to the fan, with the lower fins (closest to the CPU) getting the least amount of flow. Therefore the heat has to transfer through the fins before it gets to an area where there is enough air flow to actually aid in heat removal. But, with the air being blown down, through the fins, there is enough back-pressure to allow the air to travel almost equally across all fin surfaces before exiting, carrying a larger amount of heat with it. Don't know if that's exactly how it works, but it made sense to me, and would explain why most newer heatsinks have the air blown through rather than pulled through the fins.

Yeah, right couldn't have said it better myself.

Me, and I've done it. I'll fix almost anything (for a price). The customer originally used vegetable oil, which started to leak and smelled rather bad. Vegetable oil washes off easily with soap and water. I just drained the oil into a large pan and gave the computer a soap and water bath with automile car wash nozzle/mixer. I then blew away as much of the residual water with a compressed air hose (especially under the board components) and let it dry for about 2 days. I did some upgrades, tested the machine with a large fan blowing cooling air, and eventually refilled it with mineral oil. I had to reach inside and fish out some loose screws (oops), but that was easy enough wearing rubber gloves.

Admittedly, it's a mess, but there was sufficient justification for this customer. He wanted a totally quiet high powered computer and was more than willing to deal with the complications.

I've bought about every heatsink or fan imaginable, given and within, as another mentioned, an axiomatic engineering construct concept -- 'if it's done right [in the first place], it's time to [attempt] an improvement, [if and while not in need of entirely new construction concepts]';- time simply follows, technologically speaking, to march upon and then past any R&D Dept., in failure aptly to communicate, if at least not uniquely, then an underlying implementation of adequacy pertaining to key hard and software, constructional elements coming in, daily, across so broad a field as computers. Where, specifically, I see you for a fit is at a coincident juncture of heatsink fins and augmented airflow, both being common terms to common acceptance for pragmatic practice. The argument you have placed to ally yourself as well follows true to the same axiom: that being one, effectively, of a quest for perpetual motion: for so long as the key component of design efficiency is established, widely employed to a common basis of underlying industrial acceptance, the cost factor, then, is effectively one which requires of me nothing more, out of my pocket, than to advance a better sense of encouragement that such benefit, you have chosen to propose, indeed, is of worthier consideration. Stop by anytime if you'd like to talk, John. My office is at the top of the stairs.

Using refrigerated air for heat sink cooling is ridiculous. The problem is that dry air is a better insulator than conductor of heat. Liquid cooling is far more efficient at moving heat than chilled air. If you're going to be moving refrigerant around, you might was well eliminate the intermediate step of cooling the air, and just cool the CPU heat sink directly.

At a previous employer, we experimented with water mist cooling of transmitters and NMR amplifiers. In effect, it was air cooling with a fine fog of water added. Cooling efficiency was spectacular. For a grossly inefficient Class A amplifier (necessary for ultra low IM distortion) burning about 500 watts of heat, the cooling air flow necessary with misting was about 1/10th. This was not cooling by heat of vaporization as the heat sinks never became hot enough to evaporate all the water. The result was a constantly growing puddle of water near the exhaust port, where the water fog hit the comparatively cold exhaust grill and condensed to a liquid. The exhaust mist also retained all the heat produced by the power amp. Put your hand in the exhaust air flow even several meters away, and you will get a nasty burn.

With todays fluidic control systems, I'm fairly sure I can design and build a mist type cooling system for computahs. The big benefit would be far less air flow necessary. However, that's the only advantage as all the other obvious problems (cost, condensation, corrosion, humidity, etc) will probably make it unusable.

.highlandtechnology.com=A0 jlarkin at highlandtechnology dot com

Doesn't lowest theta occur at MAXIMUM airflow? Think of it as a stiction layer of air. All the heat from the metal must travel through that stiction layer [which by definition is an insulative sheet of air on the heat sink.] Now as air moves rapidly over the HS's surface; that stiction layer becomes thin and voila! lowers the theta. That's why all the HS curves of temp ris for given wattage always asymptotically approach some value as the air speed increases. It's because the stiction layer can be thinned, but not removed.

(...)

Incwww.highlandtechnology.com  jlarkin at highlandtechnology dot com

Volume is good, sure, but everything else equal,that takes a bigger fan. The question is, given a fan optimize the resistance. It's an impedance matching sort of question. Constrict the air too much, with heatsink blades and the airflow goes down. Open it up and there is no contact between the heatsink and air. As others have mentioned, the point isn't to reduce the boundary layer by increasing velocity, rather to upset the boundary layer with a turbulent flow. ...just enough turbulence to upset the boundary layer but not so much as to restrict flow. Just enough heatsink material to transfer heat and not so much as to restrict flow. It's not just a single variable equation. The heat-transfer people at IBM (sat down the hall from me, moons ago) used our electronics simulation programs to design these things. Their sim models were just as large as ours and took just as much CPU time (hours).

Air has a very low specific heat. Liquids are much better. Phase change is even better but you still have to dump the heat somewhere.

How about out the drain connection? ;-)

Incwww.highlandtechnology.com  jlarkin at highlandtechnology dot com

If you have a veriable speed fan, sure, more air flow cools better. That's not what I'm talking about.

If you have some fan running full blast, with some flow-backpressure curve, and you blow it at/through a heat sink, you could vary fin density, fin shape, ducting, stuff like that. Maximum air flow implies zero interaction with the heat sink fins, which will make for zero cooling. 100% blocking by fins is zero air flow, also no cooling attributable to the fan.

Think of it as a

The other thing going on is that the thermal resistance of the heatsink itself, resistance along the fins or pins, starts to isolate the extremes of the heatsink at high heat flows. You could attach the tips of the fins to a zero-theta block and still have the thermal resistance of the fins themselves. In real life, the baseplate lateral theta gets to be a problem too; you get a hot spot at the chip or mosfet or whatever, but the sink is cooler farther away. Heat sink extrusions are apparently specified with uniform heating of the baseplate.

For a given chip, there would be a finite theta if it were bolted to the face of a half-the-universe-sized block of copper.

Are you really John Fields?

IBM mainframes all used chilled air for the peripheral components, like memory, power supplies, and controls; anything that was card-on-board technology. Smaller systems used chilled air from under the floor (heat exchanger not built into the system).

Not enough to matter. The expansion goes as absolute temperature.

That's why the exhaust port for a heat removal

There's no advantage to making an inlet port bigger than the fan that site there. Disadvantage, if it would let air leak out. But an outlet port restricts flow, so make it big.

Indeed, and add to that the fact that fans are not "resistive" air sources. The peak power point (pressure * flow) occurs at a pressure of about 25% of maximum (fully blocked) pressure. You can't operate very far from this condition or your flow will be too slow.

If you include dynamic pressure (mass flow), fans are even more nonlinear. Next time you have a squirrel cage type laying around, hook it up and play with it. Put your hand over the outlet. You'll find the velocity is great until about 1-2 diameters away, where you start feeling the force of ram air. Within about 0.5 to 0.25 diameters, pressure is maximum, because flow hasn't gone to zero yet, meanwhile static pressure is building. Put your hand all the way up to block it, and static pressure goes to maximum, but velocity goes to zero, so the power you're feeling drops sharply.

BTW, I use the example of a squirrel cage because they provide more pressure, making a more illustrative example. Regular axial fans do as well, and manufacturers typically provide comparable graphs.

Tim