I've read the paper now, so I can chime in on an informed basis.
They claim a 0.2°C/W, which would be really something. You can get down to 0.37°C/W[1] with air cooling, using heroic measures, and blisteringly awful efficiency. Doing 0.2°C/W would be a real step up.
I don't really think they've done it, here. Their experimental setup used six 1"x1" 10 watt heating elements. This is because their heatsink needs a very large cross-section to overcome the lousy thermal conductivity of the air gap between the impeller and the base plate.[2] Total area, 38.7cm^2, total power, 60W.
The Intel Core i7's heatspreader has a surface area of ~20.25cm^2, and a thermal design power of 130W. 52% smaller, 216% times the heat output. That's about four times more heat per square centimetre.
The smaller the heat source, the longer the average thermal path between the source and the air/heatsink boundry, the worse the C/W, and the less effective the heatsink will be. If they had actually used a computer processor, rather than a bunch of heating elements, then my WAG is that they would have done 0.35°C/W.
It's a beautiful idea, but their setup looks nothing like reality.
2: They say that, due to the high sheer speed, there's no boundary layer, which "increases thermal conductivity several-fold". Well that's a cool story, bro, but air (0.025 W/(mxk)) is still sixteen thousand times less thermally conductive than copper! (401.0 W/mxk)) If you increased the thermal conductivity of air by 6.4 times, then it would be as conductive as... rubber, something which is not world renowned as a good conductor of heat!
Dan's article is very old (2003). At that time is was not easy to mount a big cooler on a motherboard. The heatpipes were not widely used.
These days you can find huge coolers with (or at least claimed[1]) less than 0.2°C/W. Most of the lower power CPUs (40-50W) can run fanless at decent temperatures with half a kilo of metal fins + some heatpipes, and you can mount that kind of heat spreader easily with bolt-through screws.
Also, that metal fan is dangerous! If you ever touched even a slow-spinning plastic fan's blades, you know how painful (or bloody) it can be. And that was only a very light plastic blade, probably less than 30 grams. Now think about the momentum of a 200 gram metal thingie spinning with more than 1000rpm (as I understand, they tested it to 7-8000 rpm). You really want to be sure you will not touch it while spinning. But if you enclose it, even in a wire cage, the performance will decrease.
I would hesitate twice and then hesitate one more time before placing any confidence at all in a manufacturer supplied number for a statistic like C/W, which can be so easily fudged, as we have seen.
But yeah, I've been out of the overclocking scene for a couple years, now. I have no idea how well heatpipe heatsinks perform, other than "better".
As for the danger of a large piece of metal spinning at kRPM, I agree. This is bad enough that there probably would have to be an safety interlock on computer case access panels that reverses the impeller to a halt when the enclosure is opened.
SPCR is a review site pretty serious about their methology; their newest heatsink reviews don't have °C/W figures due to difficulty measuring the power output of socket 1366 CPUs, but here's a mid-2009 review of a huge heatpipe/tower heatsink: http://www.silentpcreview.com/scythe-mugen2
Page 6 states °C/W values of 0.29 down to 0.14 depending on the amount of air pushed with the fans. SPCR's test fans are pretty low-flow to begin with (47 CFM/1080 RPM at 12 V), so you could probably go slightly lower by using real screamers.
Could diamond-like carbon (which is very slick) be used instead of the air bearing, so that the spinning heatsink remained in contact with its heat source?
You raise a very good point. The philosophy thus far is to see how far we can get with conventional, inexpensive materials, and periodically reevaluate where problems remain. Low-cost manufacturability is critically important. For example, in the case of the rotating heat-sink-impeller we are converging towards cold-forging as the fabrication process. Having said that, in a manufacturing setting many seemingly exotic coating materials are used on a routine basis in small quantities without incurring significant additional cost (e.g. CVD coatings for tungsten carbide end mills). With regard to CVD diamond coating, the main questions I have concern the current state of affairs in this technology area. How expensive are such coatings, and what are typical values for coefficient of friction?
With regard to the thermal resistance of the air gap, unfortunately it is not possible to reduce the air gap much below ~10 microns without incurring large frictional losses; the power dissipation associated with shearing of the “fluid” layer in the air gap regions scales as 1/h, were h is the air gap distance. The approach you suggest would be tractable at very low pressure, but at very low pressure convection cooling is ineffective.
Further reply from study author in response to my message stating that I had posted his reply to this forum, and pointed him to it:
===
Thanks for your interest. With regard to filling the air gap region with conductive fluid, the problem is that frictional shearing losses become prohibitively large even at low rotation speeds. It’s not that it wouldn’t work at all, but that it wouldn’t work very well; thermal conductivity is important, but so is viscosity.
There a number of other issues that apparently have caused confusion as well. Maybe we can use this forum to provide clarification. Please feel free to post what follows.
It appears that many people were unclear about what I was attempting to convey with regard to the subject of dust fouling. I did not mean to imply that there is literally no dust fouling; some dust accumulation eventually becomes visible to the naked eye on the very leading edge of the blades. The point is that dust fouling is reduced to such a large extent that we are unable to detect any degradation of cooling performance operating the device in a relatively dirty environment over an extended period of time. Thus for all intents and purposes the dust fouling problem has been taken of the table. In contrast, with conventional CPU coolers, eventually the entire heat exchanger surface becomes entombed in dust.
Some people have expressed concern that such a rotating heat-sink-impeller would constitute a safety hazard. Any real world device would include a screen, grill, or other form of protective enclosure. Such protective measures are widely used on conventional fans. We photographed our device without an enclosure so that it would be easier for people to see what it looks like.
There seems to be confusion about where the potential for significant electricity savings resides. The vast majority of it is associated with applications such as air conditioning and refrigeration, not electronics cooling. But such energy sector applications will only materialize if air bearing heat exchanger technology proves amenable to size scaling. We are in the process of evaluating this question.
Many have expressed skepticism about the practicality of a 0.001” air gap. That’s certainly understandable, and it was one of the first things we investigated. After all, if the requirement for a small air gap precludes the possibility of low-cost manufacturing, reliability, etc. then I would be the first to agree that all of this is a pointless exercise. The end-of-project report has a discussion of why this is not the case. Another counter-intuitive point is that air bearings are extremely mechanically stiff, rugged and reliable. This is also discussed the report.
On a related subject, many were concerned about the manufacturability of the heat-sink-impeller. We are converging on cold forging as the best route to low cost fabrication. We understand that if we can’t drive the cost down that air bearing heat exchanger technology will have little impact.
Others have pointed out that intuitively it would seem like the last thing you’d want to do is intentionally introduce an air gap (rather than something like thermal grease) in the thermal conduction path between the CPU and heat exchanger. Qualitatively this sounds like a persuasive argument. But as discussed in the report, quantitatively, the numbers (gap distance, gap area, thermal conductivity of air, enhancement of conductivity by convection) work out quite well. For a 10 cm diameter device, an air gap resistance of 0.02 C/W is certainly feasible.
Another important point is that the version 1 prototype device discussed in this report is badly unoptimized. The main objective for version 1 was to test our hypotheses regarding the advantages of such a device architecture. If things continue to go well in lab, I suspect eventually we’ll end up at about 0.05 C/W for a 10 cm diameter device that’s of order 3 cm high, operates at three to four thousand rpm, and consumes about 5 watts of electrical power. But believe it when you see it. There’s always risk involved in try to solve tough problems. We’re giving it our best shot. We’re also working on alternative device geometries that may be capable of providing considerably better performance.
What else? A couple people stated that the thermal brick wall is at 4 GHz, not 3 GHz. Fair enough. My point is that if you introduce a drastic improvement in thermal management technology, whatever the number for thermal brick wall may be, it gets pushed a lot higher.
Other people had questions about why such a heat-sink-impeller is so quiet. What it boils down to is that you’ve got a lot more flexibility with regard to blade geometry than you do with a fan. That means you are free to design the blade geometry to smoothly split and smoothly rejoin the flow field at the impeller entrance and exit; the device architecture allows you to decouple the engineering constraints of adequate air flow and low noise.
A couple of people surmised that the air bearing heat exchanger requires a source of compressed air because we used a hydrostatic air bearing in the version 1 device. As described in the report, in a real-world device you’d use a hydrodynamic (or self pressurizing) air bearing. That’s what we’re using now in versions 2 and 3. The use of a hydrostatic bearing in version 1 was an experimental convenience.
I have to go to a 5:30 meeting, so I will leave it at that. Hopefully people will find portions of the above material informative.
Sincerely,
Jeff Koplow
Sandia National Labs
Livermore, CA
What I especially like about this reply is how it clearly shows how easily he can answer the questions and objections raised. It acknowledges they are reasonable questions and as such they have already been considered. There is a current answer, that may change if new ideas/evidence come to light.
Readers of this kind of news often seem to underestimate the kind of thought that went into an idea/invention/prototype. They go "well, that won't work, because ..." and they come up with some reason that may sound reasonable (or simply is reasonable), but just isn't important enough. I think it would be good if more scientists replied in this kind of matter to comments to news of their ideas/inventions. It would, I hope, increase the respect people have for the work of scientists.
Adding DLC doesn't change the problem of connecting to the heat source. There is still an air gap.
A conductive fluid bearing might work better than an air gap. DLC vs copper makes little difference since that's not the problem.
I see you just edited it to say that DLC is very slick - that doesn't help as much as you think. It's a fact of heatsinks that two solid surfaces only touch in three places. The rest is filled with a tiny air gap. You typically fill in this gap with a fluid.
If you have two completely rigid objects, then as soon as they touch on two points, those two points define an axis. Rotate the objects on this axis until they touch at a third point. Now they are fixed relative to each other, and you can't move them to touch at more points. If you slide the objects on each other, as soon as they touch at a "new" point, one of the other three will be slightly further away, and won't touch any more. Unless you use a flexible material, or somehow make a perfectly flat surface, they will only touch at three points.
(page 18 of PDF) "The prototype device is configured as a static (externally pressurized) thrust bearing. In realworld thermal management applications such an externally pressurized air bearing would be replaced by a hydrodynamic (self-pressurizing) air bearing, which uses a minute fraction of the mechanical power supplied by the brushless motor to generate the required lifting force."
So the external pump at the moment keeps the big rotor above the surface? They haven't demonstrated how the setup would otherwise work, but maybe for "self-pressurizing" variant the RPM would have to be really high, which would again mean not quiet at all?
So lets say their device can handle 60W when it is 38.7cm^2 (or a bit more than 1.5W per square cm. Now you want to cool a 150W processor. That means 100cm^2 of cooling surface. If I did the math right 100cm^2 is sqrt(100/pi) * 2 diameter circle. Or 12cm diameter circle. So lets call it a 4 3/4" diameter circle with one of these gizmos on it.
That isn't a huge stretch to imagine. I'm thinking a copper slug which sits on the processor and makes this 4.75" diameter 'landing pad' for their cooler. Doesn't work well if the processor isn't oriented horizontally though.
In data centers this might up reliability and density if the net solution was 'shorter' than the current passive sinks.
It's common in high-end CPU HSFs to mount the heatsink above the copper slug on the CPU, connected with heatpipes - a setup that would work well for a cooler of those dimensions.
They claim a 0.2°C/W, which would be really something. You can get down to 0.37°C/W[1] with air cooling, using heroic measures, and blisteringly awful efficiency. Doing 0.2°C/W would be a real step up.
I don't really think they've done it, here. Their experimental setup used six 1"x1" 10 watt heating elements. This is because their heatsink needs a very large cross-section to overcome the lousy thermal conductivity of the air gap between the impeller and the base plate.[2] Total area, 38.7cm^2, total power, 60W.
The Intel Core i7's heatspreader has a surface area of ~20.25cm^2, and a thermal design power of 130W. 52% smaller, 216% times the heat output. That's about four times more heat per square centimetre.
The smaller the heat source, the longer the average thermal path between the source and the air/heatsink boundry, the worse the C/W, and the less effective the heatsink will be. If they had actually used a computer processor, rather than a bunch of heating elements, then my WAG is that they would have done 0.35°C/W.
It's a beautiful idea, but their setup looks nothing like reality.
1: http://www.dansdata.com/quickshot012.htm
2: They say that, due to the high sheer speed, there's no boundary layer, which "increases thermal conductivity several-fold". Well that's a cool story, bro, but air (0.025 W/(mxk)) is still sixteen thousand times less thermally conductive than copper! (401.0 W/mxk)) If you increased the thermal conductivity of air by 6.4 times, then it would be as conductive as... rubber, something which is not world renowned as a good conductor of heat!