Flexible heat pipes: the future of PC cooling?

[DrMrLordX]

Check out this paper from 2013:

http://www.colorado.edu/MCEN/mems/docs/Flex TGP Paper JMM.pdf

If you look at section 4: thermal performance evaluation, you'll see that the flexible heat pipe evaluated exhibited thermal conductivity ~4.6x better than solid copper. Performance would be inferior to a solid copper heatpipes as we're used to seeing in heatpipe-based HSFs, but still . . . can you imagine the possibilities of a flexible heatpipe array?

 

[BonzaiDuck]

Check out this paper from 2013:

http://www.colorado.edu/MCEN/mems/docs/Flex TGP Paper JMM.pdf

If you look at section 4: thermal performance evaluation, you'll see that the flexible heat pipe evaluated exhibited thermal conductivity ~4.6x better than solid copper. Performance would be inferior to a solid copper heatpipes as we're used to seeing in heatpipe-based HSFs, but still . . . can you imagine the possibilities of a flexible heatpipe array?

I can imagine it contributing to further developments. We'd seen all sorts of innovations coming on the market: first, tower-coolers; then, "direct-touch." Part of the problem as I see it is the number of pipes you can array on a socket-1150 or 1151 processor. Someone must have experimented with piling them on top of each other, and it might not have proven more effective. I don't recall ever seeing two layers of pipes welded into a single heatpipe base.

I may have been examining water-cooling options for many years without actually doing it. It's a balance between cooling-benchmarks, avoidance of complexity (with multiple failure-points) and space-in-the-case. You either commit vented case surface area to a radiator, or you choose to fill your case with heatpipe fins. So compactness of a heatpipe cooler is an ongoing issue, no less than the prospect of putting a 360mm radiator in the case front or top panel. But the water-cooled options are not as flexible for case choices -- in some respects, and large air-coolers are equally less likely to fit in small-form-factor cases.

One innovation I remember was the expensive Zalman computer case, which (if I remember correctly) came with heatpipe "assemblies" that secured one end to a component and the other to a case panel. The case was touted as "fanless." I'm comfortable knowing that I didn't shell out the $200 to $300 for that case.

 

[MongGrel]

Sounds like it is more trouble than a solution to me.

Reinventing the wheel ?

 

[DrMrLordX]

Think of it this way: we use water cooling to connect heat sources to radiators that are (generally) too large to connect directly to the heat source itself via a "traditional" HSF or an HSF with rigid heatpipes.

We generally can't use rigid heatpipes in the same application thanks to their rigidity; namely, if we tried, we'd be largely foiled by the fact that our intended heat source was positioned in different physical locations from one PC to the next (different socket positions, etc).

Flexibile heatpipes should allow us to connect a humongous radiator to one or more remote heat sources without the use of water, pumps, reservoirs, or any of the other trappings of a water cooling solution. You could have an entire case made of radiator elements with flexible heatpipes linking the CPU, VRMs, and possibly GPUs to the "case". Obviously making sure that you'd have enough heatpipes from any one heatsource to the rad would be an issue . . .

 

[VeryCharBroiled]

Flexibile heatpipes should allow us to connect a humongous radiator to one or more remote heat sources without the use of water, pumps, reservoirs, or any of the other trappings of a water cooling solution. You could have an entire case made of radiator elements with flexible heatpipes linking the CPU, VRMs, and possibly GPUs to the "case". Obviously making sure that you'd have enough heatpipes from any one heatsource to the rad would be an issue . . .

that sounds great. removes the multiple points of failure that water has, is maintenance free, and the radiator placement and pipe routing is limited only by your imagination.

 

[DrMrLordX]

Right, that's what I was thinking. The concept doesn't seem to have caught on yet, though there is one commercial product that appears to be based on it:

https://www.furukawa.co.jp/review/fr025/fr25_14.pdf

Interesting that it's flat like a ribbon.

 

[BonzaiDuck]

Think of it this way: we use water cooling to connect heat sources to radiators that are (generally) too large to connect directly to the heat source itself via a "traditional" HSF or an HSF with rigid heatpipes.

We generally can't use rigid heatpipes in the same application thanks to their rigidity; namely, if we tried, we'd be largely foiled by the fact that our intended heat source was positioned in different physical locations from one PC to the next (different socket positions, etc).

Flexibile heatpipes should allow us to connect a humongous radiator to one or more remote heat sources without the use of water, pumps, reservoirs, or any of the other trappings of a water cooling solution. You could have an entire case made of radiator elements with flexible heatpipes linking the CPU, VRMs, and possibly GPUs to the "case". Obviously making sure that you'd have enough heatpipes from any one heatsource to the rad would be an issue . . .

I had an "air-cooling" epiphany with a CLU-relidded Skylake. At the same clock speed, I can beat an EXOS dual fan external water cooler by 5C if it is cooling a retail-box un-de-lidded Skylake.

Among heatpipe coolers, AiOs and comparable custom kits, I didn't even decide against going with a Predator 240 or H 240 X2 until I was able to cross-reference some reviews to find the ThermalRight LG Macho within 5C degrees of the Predator. The CLU-relid buys me 12C degrees.

The Macho seems more compact than other coolers with similar performance and size. It's actually lighter than the NH-D15, but it sure does leave the motherboard and RAM accessible.

I just don't know what innovations we'll find in this area within a year or two.

But someone here already got it right. If you get the same results with an air-cooler, it eliminates a considerable handful of failure points. I don't know if I share the vision though, with flexible pipes connected all over the case interior. And that's no criticism: it's just about "shared vision."

Sooner or later somebody will figure it all out. In the meantime, I'm miniaturizing my front panel devices, using 2x 140mm fans in the front, and I'm all ready for a 280mm radiator of more than adequate thickness. And if I'm not so inclined later in the game, this is pretty damn good. LinX, 4.6 Ghz, ~70C package temperature. Is this great or what?

 

[DrMrLordX]

That is good. I'm pretty happy with my nh-d14 for now, but eventually I am going to want a bigger set of fins, and stuff like the Macho and nh-d15 aren't that much bigger or better.

Plus I think flexible heat pipes would have a lot of applications in the server room.

 

[BonzaiDuck]

Not something I can argue to the contrary.

I'm still running an NH-D14 on a 2600K system being prepared for shipment and sale to a friend. I think the Macho performs just a tad better than the D15. But it takes up less space in the case.

What happens with this -- one invests thoughtfully in cooling before finding the limits. We anticipate from gathering facts just about what to expect. In this, I've become "better" or so I think. With some exploratory overclocks, you then realize that you'd rather keep load voltage below 1.38V. And you discover that your cooling is so good, that it almost seems excessive.

But since it's not chilled water, it's just a good air-cooling solution on a CLU-relid. The objective is really to "cool the processor" -- so who cares how you do it, as long as you get the results you want?

 

[Valantar]

I agree that this is an interesting concept (I've pondered DIY heatpipe bending for passive cooling several times), but I can't see it have much of an application in PCs outside of custom low-power builds. From looking extremely briefly at the paper, the implementation doesn't seem to lend itself to high heat output uses like CPUs or GPUs - the prototype is large in area (far wider than a CPU), yet fails to properly transfer heat when exposed to more than 18W of thermal load. In other words, you'd need 3-5 of these to effectively move the heat output of a desktop CPU. Their physical properties also fit poorly with integration into a heatsink, as a wide, flat polymer surface like this would both be hard to bond to a heatsink and would interfere with airflow unless oriented perpendicular to it - requiring a huge amount of space. Also, the polymer surface makes for inefficient thermal transfer from a heat source (how do you ensure a flat and even contact area with the CPU?), and both this and the size prevents multiple "soft heatpipes" being used on the same CPU (stacking vertically would be very, very inefficient).

I wonder if they could make these "taller" and wider, i.e. more akin to the cross-section of a regular heatpipe. That would at least improve usability, if not fix every problem with them.

 

[DrMrLordX]

If you look at the commercialized product from Furukawa (https://www.furukawa.co.jp/review/fr025/fr25_14.pdf) you can see that they're extremely thin. I'm not sure how Furukawa recommends that their Pera-flex product be bonded to metal for proper heat transfer. But with such a tiny amount of thickness, it would be pretty easy to bond a dozen or so of the things to a copper block. Modern HSFs already have 6-8 heatpipes configurations anyway. All heatpipes have their limits.

I do think that the Furukawa product would be ineffective in an HDT configuration.

 

[Valantar]

If you look at the commercialized product from Furukawa (https://www.furukawa.co.jp/review/fr025/fr25_14.pdf) you can see that they're extremely thin. I'm not sure how Furukawa recommends that their Pera-flex product be bonded to metal for proper heat transfer. But with such a tiny amount of thickness, it would be pretty easy to bond a dozen or so of the things to a copper block. Modern HSFs already have 6-8 heatpipes configurations anyway. All heatpipes have their limits.

I do think that the Furukawa product would be ineffective in an HDT configuration.

I see they use a metal foil container rather than polymer, which should bode well for thermal transfer. However, stacking these would lead to rapidly diminishing returns - after all, each layer would effectively insulate the next from the heat source. After 3-4 layers, I'd expect this to become quite inefficient.

Also, note that their example is based on a 4W 20*10mm heat source. Sure, that's a lot smaller than a CPU, but (given ideal area scaling and no added thickness) a 40*40mm cpu would need a thermal output of just 32W or less to maintain thermal transfer as they describe it.

Now, we don't know anything about neither area scaling, potential thickness increases, or stacking, so it's hard to tell either way - but I'm skeptical of practical applications outside of smartphones and tablets.

Also, the product as demonstrated is only 150mm long - far too short to lead heat from a CPU to for example a chassis fan mounted radiator. Making it longer might be easy, or have serious consequences for heat transfer. We simply don't know.

The flatness of the "heatpipe" still remains a problem, though, as we still have the problem where only one can directly contact the CPU, unlike traditional rigid heatpipes where a bunch can run side-by-side. Not to mention that the non-conducting edges of the "pipe" would have to extend outside the edges of the CPU, making mounting hardware large, complicated and difficult to implement.

Sure, modern HSF coolers have 6-8 heatpipes, but these all contact the CPU equally (directly or through a base), an implementation that would be impossible with a flat product like this. Thus your comparison is invalid.

This is all without discussing how this would interface with a heatsink - this would be even more complicated than interfacing with a heat source. You could fit a traditional flat-bottomed fin-based heatsink directly onto one of these, but these are notoriously inefficient. You could layer several of these with heatsink fins like water channels in a liquid radiator, but this would require some interesting bends to work (in addition to efficiently spreading heat across several of these pipes). You could use fixed heatpipes to lead the heat further into a heatsink, but that just seems silly. The best use of these seems to be from one flat surface to another, so from a low-power CPU to a finned case side panel seems like a good implementation.

I'm still struggling to see how this would work with a full power desktop CPU in anything resembling a desktop case.

 

[DrMrLordX]

I think it could be done . . . it would represent some engineering challenges getting it to work, but I'm fairly certain that you could bond a number of these to a copper base similar to a non-HDT heatpipe configuration and get some decent heat transfer coming out of those things.

Length of the pipe should not be a big issue. According to the literature I've read on traditional heatpipes, effective k value increases as the pipe lengthens to represent the fact that the wicking effect is uniformly useful regardless of the length of the pipe (within certain limits). Pipe bend and orientation would be a bigger problem.

 

[Valantar]

I'm fairly certain that you could bond a number of these to a copper base similar to a non-HDT heatpipe configuration and get some decent heat transfer coming out of those things.

You keep saying this, but you don't seem to take into account the radical difference in the physical dimensions of these and traditional heatpipes and the huge difference this would make in mounting. Can you explain exactly how you would bond several wide, flat objects of a fixed size to another flat, roughly square object of a similar fixed size - with all of them touching equally? This just isn't physically possible - unless the CPU IHS is a Tardis. If the "pipe" is 40mm wide, then only one can directly contact a 40mm wide CPU IHS. It's really that simple.

A round, rigid heatpipe with 8mm diameter (flattened on one side or attached to a base plate) can be stacked side-by-side with four other identical pipes to cover a 40mm wide CPU IHS. With a flat, 20mm wide flexible heatpipe (like the 4w capable 0,7mm thick one in the linked Furukawa paper), the number it's possible to attach in one layer to the IHS is two. That's 60% less. Not to mention that 8mm is larger than what's used in most CPU coolers - with 6mm round pipes, the number is suddenly 7 (with 2mm overhang). Add to that that they freely admit that these are worse at thermal transfer than standard rigid heatpipes, and you're in trouble.


Now, pardon the awful ASCII illustration:

Normal heatpipes:
 ___  ___  ___  ___  ___
/   \/   \/   \/   \/   \
|_1_||_2_||_3_||_4_||_5_|    <--- 5 heatpipes
|----------cpu----------|

Flexible heatpipes:
==========================     <---1 heatpipe
|----------cpu-----------|

As I said, we don't know how these scale in thickness. Retaining flexibility would probably be a challenge, but if they could make these 4-6mm thick, that would make a huge difference. Given linear thermal transfer scaling, that would bring the thermal transfer properties of a 4x40x150mm "pipe" (twice as wide as the Furukawa one, same length, just thicker) from 8w to 46w. Which would make for a decent ULV CPU cooler. If a second layer was added and retained, say, half the thermal transfer (remember, the heat has to pass through the first layer of pipe to reach this one - I have a feeling I'm being generous here), that adds up to 69w - enough for an i7-6700. Barely, though. And if the first "insulating" layer dropped the possible heat transfer by half, that would continue - i.e. 3 layers would be 80w, four would be 85w, and so on. Now we're nearing a stock-clocked 6700K - with four 4mm heatpipes stacked. And these (fat, wide) heatpipes would need to go somewhere to dissipate that heat. Sure, you could thermal-epoxy them to a big case panel (hopefully with fins on it). Otherwise, you still have the issue of how to efficiently mate these pipes to a suitable heatsink.

 

[DrMrLordX]

They're not all going to bond directly to the IHS. Like I said, HDT wouldn't work with those. If you've seen any of the larger HSFs out there, like an NH-D15, you'll see that the copper base is a good bit larger than the IHS of a typical consumer-level CPU.

The easiest configuration would be to go back to an old-fashioned desktop style case (read: not a tower) and mount a big rad on top of it. Then you run pipes vertically downward from the rad through the case lid and into a block roughly the size of the base of a large HSF with appropriate mounting hardware. Since it's flexible, you can reposition it within reason (the paper says 90 degree bend is the best they can get out of it).

The only downside I see to that configuration is that only a small portion of the pipe would actually have contact with the base, so thermal transfer wouldn't be as good as if the entire pipe looped through the base. You'd have to make up for it in volume of pipes. And if you wanted to get fancy you could go with a slightly larger base and replace the solid copper with a vapor chamber, though those can be a bit expensive. At that point you could potentially consider horizontal mounting, putting the rad on the back of (or offset to the side of the back of) a traditional tower case.

Use of a vapor chamber would also negate the insulation effects of stacking pipes.

Distributing heat from the flat pipes to fins in the rad would also be fairly easy. You'd run them to another copper block (or vapor chamber) and run traditional round heatpipes out of that up and down the fin stack.

 

[Valantar]

They're not all going to bond directly to the IHS. Like I said, HDT wouldn't work with those. If you've seen any of the larger HSFs out there, like an NH-D15, you'll see that the copper base is a good bit larger than the IHS of a typical consumer-level CPU.

The easiest configuration would be to go back to an old-fashioned desktop style case (read: not a tower) and mount a big rad on top of it. Then you run pipes vertically downward from the rad through the case lid and into a block roughly the size of the base of a large HSF with appropriate mounting hardware. Since it's flexible, you can reposition it within reason (the paper says 90 degree bend is the best they can get out of it).

The only downside I see to that configuration is that only a small portion of the pipe would actually have contact with the base, so thermal transfer wouldn't be as good as if the entire pipe looped through the base. You'd have to make up for it in volume of pipes. And if you wanted to get fancy you could go with a slightly larger base and replace the solid copper with a vapor chamber, though those can be a bit expensive. At that point you could potentially consider horizontal mounting, putting the rad on the back of (or offset to the side of the back of) a traditional tower case.

Use of a vapor chamber would also negate the insulation effects of stacking pipes.

Distributing heat from the flat pipes to fins in the rad would also be fairly easy. You'd run them to another copper block (or vapor chamber) and run traditional round heatpipes out of that up and down the fin stack.

So, in essence, you're talking about something ~90% similar to the HDPlex H5. Only with far lower cooling capacities, as that requires eight regular heatpipes to achieve passive cooling of a full-power (not HDT) CPU. Again, given that the flexible heatpipes are admittedly worse than those, you'd need 12-16 (minimum) - and a case large enough to give these room to attach to something. Also - the H5 exists. Without flexible heatpipes, just a bit of clever mounting.

The "keep-out" area of an Intel 115x socket is 95x95mm - this is the largest you could make a vapor chamber or other base plate without devising some intricate, wildly expensive and thermally lossy riser system to raise it above motherboard components. This of course neglects to mention that the mounting holes are within this area. But given that one could figure out a way to ignore this (say, screw in the plate from the back of the motherboard), you'd get a 95x95mm base. I'll freely admit I'm completely incapable of calculating potential dissipation of a 95x95mm base - would the length need to increase as well to scale linearly? What about thickness? But ignoring this, 95mm is 4,75x the width of the Furukawa strip, and with significantly added length (along the lines of 700mm total, at least - 95mm for the CPU, ~150mm for each side of the rise to the case panel, and then what's actually contacting the panel), I'd say expecting dissipating 95w seems ... plausible, if you allow for thickness to increase significantly. The most efficient and cost-effective way to implement this would be with a single, 95mm-wide "pipe," perhaps secured between metal plates for flatness and rigidity (I'd have the top plate curve away at the edges to avoid sharp bends in the pipe), long enough to extend to the case side panel above the CPU on both sides, and attaching the "pipe" to the case side along the entire length of the panel. Then, you'd have to embed heatpipes into the side panel perpendicular to the direction of the flexible "pipe" along the full length of the side panel to spread heat out, and then have the entire panel shaped into some sort of fin array. Given the size of the fin array of the H5, you could probably get away with using half an ATX case panel for the CPU given enough fins. Then it's on to the GPU - which would need at least as much dissipation again. Note that all of this requires multiple bends in each pipe, reducing heat transmission. But with enough slack, the bends wouldn't be too sharp.

I sketched this out roughly:
https://1drv.ms/i/s!Akhd5yRCmjurx2bKu55qNvBQ3eDN

And sure, this would work. But it would be far more complex than the HDPlex H5, with no significant gains. Zero. Sure, a bit easier to install. But other than that? At best, similar cooling performance. Not to mention the enormous cost.
1) a complex 95x95mm vapor chamber with integrated mounting screws and hold-down plate.
2) a flexible heatpipe of at least 95x700mm
3) a whole bunch of heatpipes integrated into a heatsink/side panel. Both manufacturing and parts cost would come into play here.

The HDPlex H5 is $288. It has, essentially, 1 out of 3 of these. In other words, a case like this would cost significantly more. Not to mention that you'd have to double up on every single component (and then some) for GPU cooling. I'd estimate $500, at least.

So:
-Possible? Sure.
-Feasible? Weeeeelll.....
-Practical? Heck no.
-Expensive? Yeah. Way expensive.
-"The future of PC cooling"? Don't be daft.