I would have though the efficiency comes from spinning the heatsink and so having a very efficient metal-air heat transfer compared to blowing air over a stationary heatsink
It's weakpoint is getting the heat from the CPU across the air gap into the spinning part
I don't see the super-thin air-gap as an issue. In fact, it is probably an asset. As an example, when we build the 1500W LED array we used a very thin insulating film to transfer heat from the LEDs to an aluminum clad circuit board. If you have surface area (the stationary disk in their case) a thin gap can be surprisingly efficient at transferring heat from one element to the other. Ad to this breaking up the boundary layer within that gap and now air becomes a "better" conductor.
There is no difference between spinning the heatsink and moving air at the same velocity across stationary heatsink fins. It's about molecules of air at a lower temperature going across a surface with at a higher temperature. N molecules per unit time "absorb" some of the heat and come of of the process at a higher temperature. The surface temperature drops by a similar amount.
What kills you in both cases is a very thin boundary layer of air that sticks to the surface and has a velocity profile that goes from zero at the surface to whatever the air velocity might be. This boundary layer isn't good for heat transfer from the surface to the moving air molecules, which takes a toll on the overall efficiency of the process.
There are techniques to help break-up the boundary layer in stationary fin heatsinks. Some break-up airflow to introduce turbulence. Others use raw velocity to break it up (air impingement cooling).
I don't know how well the rotating aluminum impeller breaks-up the boundary layer surrounding the fins. It won't eliminate it completely but it certainly could be much improved from the case of stationary fins with low-speed air going across them. If you had a heatsink with stationary fins and air moving at the same speed as in the case of the rotating impeller (by using fans, of course) you'd probably get similar boundary layer effects.
What I learned is that there really is very little magic in heat transfer. I have seen some really wacky ideas brought to market that always seem interesting but have never proven to beat the simple solutions. Funny enough, other than materials, most of the focus is always around dealing with the boundary layer, an issue that disappears (in terms of its significance) once you move into fluid-based cooling.
Caveat: I have not studied rotating aluminum impeller/heat-sinks, so I could very well be missing something that I simply don't know. My opinions are based on thousands of hours of research and FEA work in looking for solutions for the aforementioned project.
I think you failed to actually read TFA. Unlike you, I know almost nothing about heat dynamics, nor about FEA, so my comment is predicated entirely on the Sandia staff being truthful in their video (and being quoted accurately in ars's Q&A article). I claim zero authority on the subject.
Your top-level comment claims that the breakthrough in this is in the thin-air gap, rather than in the centrifugal force. Respectfully, Sandia's Jeff Koplow specifically claims otherwise, in detail.
Koplow claims the boundary layer is the key problem (a claim with which you appear to agree). He further claims that when the radiator is spinning (or otherwise accelerating) that the boundary layer thins, and that in their application it thins by roughly 10x. See the video at 1:25 or so. And while I don't trust my under-educated intuition particularly heavily, it's very easy to imagine how accelerating the fins is fundamentally different from blowing air across them, in terms of how it affects air molecules in the boundary layer. Again, note that you specifically claimed "There is no difference between spinning the heatsink and moving air at the same velocity across stationary heatsink fins", and Koplow has specifically claimed that this is not true.
The fluid dynamic bearing only becomes relevant as a secondary problem: if you're going to spin your radiator but not your heat source, obviously there's a transfer problem. Apparently this is easy enough to solve, I guess? Koplow did mention (a year ago) that they were considering adding roughness to the revolving surfaces to perturb the air in the gap to improve transfer.
Oh, and you repeatedly say "they should use fluid". Well, first of all my little pedantic nit-pick is that air is a fluid. But more interestingly, the article contains a link to Q&A with more technical details, in which that very subject is discussed! The short version as I understood it is this: viscosity kills you.
I guess what I'm saying here is this: you started off by saying the article was wrong (about the centrifugal force), and went on to question all the design choices involved, but I think you actually skipped the bit where you read the article in enough detail to know whether or not it was wrong. Even though you obviously have a much better background in the material than I or most other commenters.
This news isn't new. Articles and at least one paper came out, if my memory serves me, two to three years ago. Back then I went through the available data in detail. I didn't need to dive into the article posted to HN to know what they were doing. Still, I did read the entire article and watched the video before posting.
As for the boundary layer issue. A fluid (OK, liquid) based cooler has virtually none of these problems and does not require having MULTIPLE metal masses inside your computer spinning at 5,000 RPM (per the paper on the Sandia site).
Remember that you need to cool memory, graphics cards and other elements in a typical design. The air-based CPU cooler moves air around the CPU cabinet and out the back or top. All of it serves to cool other elements. Dust or not.
You can't put a bunch of 5,000 RPM coolers inside a
In a typical data center you have reasonably-clean air available. Dust should not really be a problem except for the most neglected portions of an installation. I have seen systems in service for years with no indications of dust accumulation at all.
Dust can be an issue in office or home environments. Even then, from personal experience (and only from personal experience) I have never seen a problem.
I do think that the Sandia heatsink might have interesting applications as part of the heat-exchanger in a liquid-cooled system.
If you have a centrifugal fan turning at 5,000 RPM you are going to move a lot of air radially out. So far all of their experiments seem to show the device working well in the context of pretty much an open air environment. You can't have this pump simply circulate hot air inside a computer cabinet. Because of that you will need to surround it with an intake structure as well as an exhaust structure. This device sucks a lot of air at 5,000 RPM. That, without a doubt, will be noisy.
There's also another element here that is not being compared. How many cubic feet per minute of air is this device moving? How would a stationary fin heatsink perform if you moved that much air through its fins.
A few years ago we modeled and built a custom heatsink that consisted of a centrifugal fan mounted at the center of a field of fins located at the exhaust of the fan. Put another way, fan mounted at the center of a flat plate, intake is at the center, exhaust is radially outward. The fins where located to "grab" and channel the exhaust flow. They were also "ducted" meaning that the top of the fins had a "roof" so no air could escape without bathing the entire fin. This heatsink performed very well. Expensive to manufacture, but it did very, very well. We could custom machine boundary layer control elements into the fins and do better yet. The fan was an off-the shelf plastic DC brushless centrifugal fan with no thermal properties other than moving lots of air. And it was quiet.
I am not necessarily putting down the Sandia fan. I am simply saying that one should be careful not to be attracted to new shiny things without a little critical thinking. I have seen companies waste millions by jumping into technologies that sounded great on paper an were later found impossible to commercialize due to a million real-world issues.
>I don't know how well the rotating aluminum impeller breaks-up the boundary layer surrounding the fins. It won't eliminate it completely but it certainly could be much improved from the case of stationary fins with low-speed air going across them. If you had a heatsink with stationary fins and air moving at the same speed as in the case of the rotating impeller (by using fans, of course) you'd probably get similar boundary layer effects.
I think the point is that it's a lot easier to get a piece of solid metal moving very quickly with little noise than to get air to do the same.
It has to get that air moving at comparable speeds to be as efficient as the traditional way. The trick here is to get the air move faster with less sound. It's possible because of larger airfoils, so they can get more laminar flow and with more laminar flow comes less sound.
Right, if you were building a wing. The problem is that here we are after heat exchange and, as I understand it, the last thing you want is laminar flow due to the accompanying boundary layer.
wow - that is a surprisingly thorough introduction to thermal transfer. Based on your experience, to increase thermal transfer would you want to dimple the surface or add protrusions? (I know, at the extreme these are almost the same thing). I was reading golf ball dynamics and not sure if it applies or not.
Both of those ideas work to varying degrees. You can "trip" the airflow at the leading edge of a heatsink fin by introducing notches and protrusions of different geometries. This will make the airflow past that point turbulent for a certain distance. Depending on geometry you might have re-attachment, which leads to rebuilding the boundary layer. Dimples and surface roughness can help too.
In the end you have to deal with the real world. Some of these techniques are really good to grab dust in the airflow. If that happens, you have, ultimately, created a problem greater than the one you started out to solve.
After all I've been trough I have become a huge proponent of fluid based cooling. I don't think there's any way to have forced air cooling even begin to compare with the potential performance gains of using liquids to mechanically move heat around.
Wouldn't the boundary layer thickness be affected by the centrifugal force that presses the air against the fins? I was under the impression that that was the main advantage of this design.
I seriously doubt it. That said, I'll claim ignorance here because I have not studied that part of things in depth. I know that it can take a lot to break-up the boundary layer in any meaningful way. In aircraft design there are reasons having nothing to do with heat transfer to want to break-up the boundary layer. As I understand it this can be difficult to achieve across a large surface (I'm assuming at sub-sonic speeds, don't know). They even resort to such devices as drilling small holes and literally sucking the boundary layer from the surface. This problem is tougher than it might seem. I am not an aero guy, I've learned a lot from thousands of hours dedicated to finding solutions to our specific problems, but I am far from an expert. Maybe an aero expert on HN can fill-in some of the holes.
At some level, think about it this way: At some point, microscopic as it might be, some of the air molecules touching the fins have to "stick" to the surface of the fins. At that point air molecule velocity with respect to the fin surface is zero. Then there are molecules that stick to these molecules one layer above, and so on. After a certain thickness the greater airflow will win out and molecules will move at the average velocity of the bulk air mass moving across the fins. What you have is a velocity profile from zero to the average air mass velocity. That's your boundary layer. You can do things to make it thinner, but eliminating it is very difficult. Techniques like impingement cooling do this to varying degrees of efficiency.
Yes, spinning the heatsink is the whole point, to take advantage of the reduced boundary effect from the centrifugal force. I imagine the force of the air pressing against the fins in this case is considerable at such high RPMS. The other clever aspects of the design like the air bearing are simply to overcome problems that result from having a heatsink that's moving with respect to the heat source.
There is of-course no theoretical difference, except this presumably gets much high air flow rates than a conventional fan + static finned heatsink. and you don't have the problem of stalled pockets of warm air in corners of a fixed fin that don't get flushed out.
One of the problems with small heatsinks is that the bulk convection flow you model at high delta-T/high power don't always work in practice with small fans and small heatsinks - this design should scale down a lot better.
My concern was that in order to get good conductivity across the air gap you would need very close tolerances which are hard to make reliably in practice on cheap consumer gear.
Also, although the fin blades themselves should clear dust - I would worry about an oil/dust/dirt film building up in the gap if it's relying on constantly forcing new air through this to make a cushion
I remember reading the paper (or "a" paper) on the Sandia cooler a couple of years ago. The author seemed to indicate that the disks spinning so-fast-yet-so-close served to break the boundary layer AND keep the gap clean and clear of dust and grime.
I'd be interesting to understand just how precisely matched the surfaces have to be for this to work well. Machining a reasonably flat reference surface on a CNC lathe or mill isn't all that difficult. The question in my mind is more about how flat these surfaces have to be. The cutting tools will leave some grooves, even if almost imperceptible. Do the surfaces have to be lapped (sanded) and polished for this heat exchanger to work well? What are the tolerances? A good shell cutter on a high-quality milling machine can produce a mirror-like surface. It's one thing to do this in small quantities and quite another in mass production (which I now nothing about).
The gap is 1-thou (30um), this isn't challenging machining - you could make the gap 30nm ! (it's a bit more annoying with copper)
Actually some machining marks would probably be good, small surface irregularities will break the boundary layer - like sharks skin and make the air flow mix better.
Agreed. I've made "reference" flat surfaces on aluminum using our Haas VF3-SS vertical milling machine with a good quality shell mill with new inserts and, if you do everything right, you can see yourself on the mirror-like surface that results. This machine is probably the grade of machine you might expect to find in a high-volume production shop.
There is totally a theoretical difference between an accelerating frame and a non-accelerating frame. And a rotating frame is accelerating, and a fan-plus-static is not.
It's weakpoint is getting the heat from the CPU across the air gap into the spinning part