News Enthusiast Develops First Low-Profile Raspberry Pi 4 Cooler With Direct-Contact Heatpipes

[Admin]

An enthusiast has developed the first low-profile cooler with direct-contact heatpipes for the Raspberry Pi 4.

Enthusiast Develops First Low-Profile Raspberry Pi 4 Cooler With Direct-Contact Heatpipes : Read more

 

[Giroro]

Putting everything in contact with the heatpipe isn't really ideal. Heat pipes are intended to transfer heat on one end of the pipe to the other, and they work because they are full of liquid.
You want everything hot on one side of the heat pipe (evaporator) and the heat sink on the other side (condenser).

Is it better with the heat pipe than without it? I don't know, probably - unless the heat load saturates the heatsink and the cpu ends up cooking everything (which I doubt would happen)... But I have a feeling it mostly won't be effective enough to be worth the effort of adding the heatpipe at all. Although as far as I remember all the chips are different heights with the CPU being tallest, so cooling the CPU by itself is most likely what people end up doing - and effectively what would happen if the CPU gets direct contact and those chips are contacting through thick cheap barely-thermally-conductive blue silicone pads.
If I were to be designing this to cool only the CPU, I would put the conductive resin at either end of the heat pipe, but a thermal insulator between the heatsink and heatpipe directly above the CPU- this would help the heatpipe to function properly.
As-is, it would be cheaper and most likely just as effective to embed a chunks of copper into the heatsink above each chip (which would also solve the issue of different chip heights) - or just make the whole thing out of copper.
A lot of pi heatsinks do have heat transfer problems, but I think part of that is because people keep selling them with thermally-insulating doublestick tape.

 

[justin.m.beauvais]

Putting everything in contact with the heatpipe isn't really ideal. Heat pipes are intended to transfer heat on one end of the pipe to the other, and they work because they are full of liquid.
You want everything hot on one side of the heat pipe (evaporator) and the heat sink on the other side (condenser).

Is it better with the heat pipe than without it? I don't know, probably - unless the heat load saturates the heatsink and the cpu ends up cooking everything (which I doubt would happen)... But I have a feeling it mostly won't be effective enough to be worth the effort of adding the heatpipe at all. Although as far as I remember all the chips are different heights with the CPU being tallest, so cooling the CPU by itself is most likely what people end up doing - and effectively what would happen if the CPU gets direct contact and those chips are contacting through thick cheap barely-thermally-conductive blue silicone pads.
If I were to be designing this to cool only the CPU, I would put the conductive resin at either end of the heat pipe, but a thermal insulator between the heatsink and heatpipe directly above the CPU- this would help the heatpipe to function properly.
As-is, it would be cheaper and most likely just as effective to embed a chunks of copper into the heatsink above each chip (which would also solve the issue of different chip heights) - or just make the whole thing out of copper.
A lot of pi heatsinks do have heat transfer problems, but I think part of that is because people keep selling them with thermally-insulating doublestick tape.

Well, the CPU is going to be the hottest component by far. Nothing else is going to heat up like the CPU will. The heatpipe in this case looks more like a heat spreader, in order to distribute the heat over as much of the aluminum cooling surface as possible. The heatpipe is going to heat up and transfer heat down its length and the aluminum is going to draw that away with its large thermal mass. This cooler might not be perfect, but it will definitely keep the Pi from overheating, probably even without any active cooling.

 

[bit_user]

Putting everything in contact with the heatpipe isn't really ideal. Heat pipes are intended to transfer heat on one end of the pipe to the other, and they work because they are full of liquid.

No, I don't believe so. A heatpipe is basically a specialized vapor chamber, and those have no real requirement on shape. The only requirement is for the "cold" parts to remain cold enough for the fluid to re-condense.

Also, heat pipes are not "full of liquid". All that's required is that you have enough active fluid for some to remain liquid within the pipe's operational parameters.

as far as I remember all the chips are different heights with the CPU being tallest,

They can include shims for the other chips, if so. Although that's less efficient (i.e. no longer direct-touch), the CPU is certainly what produces the most heat.

If I were to be designing this to cool only the CPU, I would put the conductive resin at either end of the heat pipe, but a thermal insulator between the heatsink and heatpipe directly above the CPU- this would help the heatpipe to function properly.

I think you need to spend more time reading about heatpipes and less time posting.

it would be cheaper and most likely just as effective to embed a chunks of copper into the heatsink above each chip

You don't want a "chunk" of copper. What you want is to efficiently move heat into to the fins. You don't do that with a "chunk" of anything. And heat pipes are massively more efficient than solid copper, which is why they're so popular.

just make the whole thing out of copper.

Copper is also rather expensive. I'm not sure that would be within budget.

 

[bit_user]

I just wonder if this would work better with cross-cut fins. Particularly with a hat on top, it seems like the linear fins could tend to trap hot air up against the hat.

That said, I'm sympathetic to cost constraints. A lot of people will be unwilling to use a heat sink that's near or equal to the cost of the Pi.

 

[CooliPi]

Heatpipes aren't necessary. HATs on top of the heatsink block airflow.

I've designed a different solution in mid July. Unfortunately, many hardware sites' editors are busy with new mainstream hardware and don't have time to review it. What a pity.

No ads - google my name.

 

[CooliPi]

Putting everything in contact with the heatpipe isn't really ideal. Heat pipes are intended to transfer heat on one end of the pipe to the other, and they work because they are full of liquid.
You want everything hot on one side of the heat pipe (evaporator) and the heat sink on the other side (condenser).

Is it better with the heat pipe than without it? I don't know, probably - unless the heat load saturates the heatsink and the cpu ends up cooking everything (which I doubt would happen)... But I have a feeling it mostly won't be effective enough to be worth the effort of adding the heatpipe at all. Although as far as I remember all the chips are different heights with the CPU being tallest, so cooling the CPU by itself is most likely what people end up doing - and effectively what would happen if the CPU gets direct contact and those chips are contacting through thick cheap barely-thermally-conductive blue silicone pads.
If I were to be designing this to cool only the CPU, I would put the conductive resin at either end of the heat pipe, but a thermal insulator between the heatsink and heatpipe directly above the CPU- this would help the heatpipe to function properly.
As-is, it would be cheaper and most likely just as effective to embed a chunks of copper into the heatsink above each chip (which would also solve the issue of different chip heights) - or just make the whole thing out of copper.
A lot of pi heatsinks do have heat transfer problems, but I think part of that is because people keep selling them with thermally-insulating doublestick tape.

Heatpipes can thansfer heat from one end to the other with low delta temperature. But the Pi isn't limited by too much power in one hotspot. It's limited mainly by size of the heatsink for passive cooling.

Copper and heatpipes may help with a 100W hotspot. Large area alone helps with 8W. No need to transfer a few watts a few centimeters unless you are height limited. And after an installation of a HAT on top of the heatsink, all the effort with copper and hetapipes becomes nil.

I don't agree on the last idea - many of the Pi heatsinks have a problem that they're too small for the 8W load! Or are blocked behind something, thus voiding its proper function.

Some I've seen are way way too small - just a joke. Some others are a bit bigger, but covered with fans blowing nowhere. Or with fans blowing towards each other. Or blowing through a 1mm wide hole. One another has horizontal fine fins . They must be vertical. And not so fine, because that blocks airflow.

Forced cooling is a different playground. Small fans die in a half a year. Really "good" investment... But finely spaced fins can start to work with enough airflow.

 

[syadnom]

This is a flat out bad design. Heatpipes transfer data from end to end to take heat away to a cooler spot. This puts the primary heat source directly in the middle of the pipe, so it's completely useless. This likely behaves worse the just having the heatsink.

Further, this is pushing heat from the hot CPU to other lower watt devices so will likely make the overall system less stable.

Much simpler and more effective would be a substantially larger heatsink with taller pin-fins. I understand that the desire to have a heatsink compatible with hats was the initial goal so along those lines, the heatpipe should wrap around the side to a large heatsink on the back side. The low profile heatsink on the front can be for the devices on the front, then the end of the heatpipe should be right on the CPU.

 

[CooliPi]

This is a flat out bad design. Heatpipes transfer data from end to end to take heat away to a cooler spot. This puts the primary heat source directly in the middle of the pipe, so it's completely useless. This likely behaves worse the just having the heatsink.

Further, this is pushing heat from the hot CPU to other lower watt devices so will likely make the overall system less stable.

Much simpler and more effective would be a substantially larger heatsink with taller pin-fins. I understand that the desire to have a heatsink compatible with hats was the initial goal so along those lines, the heatpipe should wrap around the side to a large heatsink on the back side. The low profile heatsink on the front can be for the devices on the front, then the end of the heatpipe should be right on the CPU.

We can get max temp delta at about 23degC with 4 core linpack load, no video output. Stock RPI4B with no case throttles at 3rd and 4th core load.

And yes, it's larger and it can host HATS without blocking airflow ;-)

CooliPi

 

[Giroro]

No, I don't believe so. A heatpipe is basically a specialized vapor chamber, and those have no real requirement on shape.

If the intention is to move heat from the 'bottom' to the 'top' of the heat pipe like a vapor chamber, then it will be better to use a vapor chamber. The capillaries in a heat pipe will interfere with that action.

The only requirement is for the "cold" parts to remain cold enough for the fluid to re-condense.

That's exactly my point. This is why it's bad design to put load across the entire heat pipe: It prevents "cold" parts from existing.

They can include shims for the other chips, if so. Although that's less efficient (i.e. no longer direct-touch), the CPU is certainly what produces the most heat.

If the CPU saturates the heat sink at a higher temperature than the other chips -which seems likely, then putting other chips in contact with it will heat them up instead of cool them.

I think you need to spend more time reading about heatpipes and less time posting.

No need to be rude

You don't want a "chunk" of copper. What you want is to efficiently move heat into to the fins. You don't do that with a "chunk" of anything. And heat pipes are massively more efficient than solid copper, which is why they're so popular.

Well the thesis of my argument is that using a heat pipe in this arrangement isn't going to transfer heat over such a small area and small load any more efficiently than plain copper. Heat pipes are massively more effective than solid copper when designed so the fluid flows properly. This one isn't designed to work as effectively as it should. I think it is adding very little, if anything to the effectiveness of the heat sink compared to solid copper. Depending on the resin holding it in there, it might not even be more effective than direct contact to the aluminum heat sink

Copper is also rather expensive. I'm not sure that would be within budget.

Heat pipes cost a lot more than the copper that they are made of. All the extra steps to inlay it into the heatsink also adds cost that wouldn't be necessary with a solid copper heatsink.

 

[Giroro]

I don't agree on the last idea - many of the Pi heatsinks have a problem that they're too small for the 8W load! Or are blocked behind something, thus voiding its proper function.

I specifically mean the pimoroni heatsink which is one of the more commonly reviewed ones
https://www.tomshardware.com/reviews/pimoroni-fan-shim-heatsink-raspberry-pi-4,6219.html

Thermal images from Tom's and others always show the surface of the heatsink as much cooler than the CPU when it's thermal throttling, with the temperature evenly distributed throughout the heatsink. This suggests there is a heat transfer problem between the CPU and heatsink.
And as Tom's pointed out that heatsink (and almost every Pi heatsink I've seen) it ships with 3M Double Coated Tissue Tape 9448A which is a "thermal tape" in the sense that it will stay sticky at high temps but is not an effective thermal conductor. 3M dosn't even list it's thermal conductivity on the spec sheet, because its not intended for that use.

But hey, I could be totally wrong, most people have never gone shopping for thermally conductive tape before, and I've never found a review where they test that heatsink with better tape.

 

[RedShiba]

Hello! Maker of the heatsink here!

When I started the design I gave myself a hard z-limit of 7mm. I added the heatpipe in order to keep the thickness of the base to a minimum and dedicate more of the 7mm thickness to the fins without loosing conductivity in the xy directions. In hind sight I should have ordered 1 heatsink with no heatpipe to compare their real world performance. It is %100 possible that the heatpipe as it is right now isn't really helping.

Another thing I forgot to mention to Ash before the article was published was that the $25 price includes the next revision of the heatsink when I finish it designing/producing it. The price of that one should be considerably cheaper and will have some neat tricks. However, setting up the site is priority one right now. (Gotta wait for my paycheck to pay the commerce fee for squarespace lol).

I've learned a lot from the feedback from both here and from reddit, It's honestly quite exiting seeing people get exited about something you've created.


-George

edit: Also yes, 8810 and 9448A tape are absolutely ass for heat conductance. 8810 was about 35c higher than cheap 1.2W/mK thermal paste from ebay.

 

[CooliPi]

Hello! Maker of the heatsink here!

When I started the design I gave myself a hard z-limit of 7mm. I added the heatpipe in order to keep the thickness of the base to a minimum and dedicate more of the 7mm thickness to the fins without loosing conductivity in the xy directions. In hind sight I should have ordered 1 heatsink with no heatpipe to compare their real world performance. It is %100 possible that the heatpipe as it is right now isn't really helping.

Another thing I forgot to mention to Ash before the article was published was that the $25 price includes the next revision of the heatsink when I finish it designing/producing it. The price of that one should be considerably cheaper and will have some neat tricks. However, setting up the site is priority one right now. (Gotta wait for my paycheck to pay the commerce fee for squarespace lol).

I've learned a lot from the feedback from both here and from reddit, It's honestly quite exiting seeing people get exited about something you've created.


-George

edit: Also yes, 8810 and 9448A tape are absolutely ass for heat conductance. 8810 was about 35c higher than cheap 1.2W/mK thermal paste from ebay.

Glad to say hello to a colleague! Salute!

 

[CooliPi]

I specifically mean the pimoroni heatsink which is one of the more commonly reviewed ones
https://www.tomshardware.com/reviews/pimoroni-fan-shim-heatsink-raspberry-pi-4,6219.html

Thermal images from Tom's and others always show the surface of the heatsink as much cooler than the CPU when it's thermal throttling, with the temperature evenly distributed throughout the heatsink. This suggests there is a heat transfer problem between the CPU and heatsink.

Not necessarily. Lack of thermal gradient at the usual load was to be expected. It shows that the heatsink itself doesn't suffer from not enough mass, but surface area.

And as Tom's pointed out that heatsink (and almost every Pi heatsink I've seen) it ships with 3M Double Coated Tissue Tape 9448A which is a "thermal tape" in the sense that it will stay sticky at high temps but is not an effective thermal conductor. 3M dosn't even list it's thermal conductivity on the spec sheet, because its not intended for that use.

But hey, I could be totally wrong, most people have never gone shopping for thermally conductive tape before, and I've never found a review where they test that heatsink with better tape.

I can even see a difference between an old white thermal paste and an old MX-2 paste from Arctic Cooling. It makes about 3degC difference. So some bad tape would certainly destroy performance and increase temperatures.

Regarding thermal performance, I use a linpack benchmark with an increasing number of threads after half an hour.

See https://www.coolipi.com/Performance.html

 

[bit_user]

after an installation of a HAT on top of the heatsink, all the effort with copper and hetapipes becomes nil.

Well, the benefit I see is that if the center of the heatsink becomes too hot, due to trapped air, then the heatpipe can still efficiently move the heat towards the edges. The heat only needs to get to a point that's cool enough for the fluid to condense.

Some others are a bit bigger, but covered with fans blowing nowhere. Or with fans blowing towards each other. Or blowing through a 1mm wide hole. ...

Forced cooling is a different playground. Small fans die in a half a year. Really "good" investment... But finely spaced fins can start to work with enough airflow.

I only seen a handful of fan-heatsinks reviewed, but the data shows that they work, however inefficient or short-lived they may be.

 

[bit_user]

This is a flat out bad design. Heatpipes transfer data from end to end to take heat away to a cooler spot. This puts the primary heat source directly in the middle of the pipe, so it's completely useless. This likely behaves worse the just having the heatsink.

A great many PC heatsinks have exactly this design, where the middle of the heat pipes contact the CPU and the ends are curled up into the fin stack. I suppose they're all bad, then?

Further, this is pushing heat from the hot CPU to other lower watt devices so will likely make the overall system less stable.

Again, you don't seem to understand how heat pipes work. As long as those other chips are hot enough to stay above the boiling point of the working fluid, then you won't see a significant amount of heat transfered into them. Most of the heat leaves the heatpipe at spots which are cool enough for the fluid to condense.

Edit: check out the videos, below. The second one directly addresses the issue of evaporating in the middle.

 

[bit_user]

If the intention is to move heat from the 'bottom' to the 'top' of the heat pipe like a vapor chamber, then it will be better to use a vapor chamber. The capillaries in a heat pipe will interfere with that action.

You assume the capillaries are directional, but they needn't be (and, in this case, probably aren't). See sintered capillaries, here:

Cheaper, mesh-type capillaries would probably also work reasonably well.

That's exactly my point. This is why it's bad design to put load across the entire heat pipe: It prevents "cold" parts from existing.

Some heat will leave the heat pipe at all points cooler than the vapor temperature. To that end, the heat pipe still serves to distribute heat into the fins. However, it seems implausible that an 8 W heat source is going to keep the entire heat pipe above the fluid's boiling point. Even without good air circulation, there'll be enough heat leaving the edges. It might even be the case that the fin block is really larger than it needs to be.

If the CPU saturates the heat sink at a higher temperature than the other chips -which seems likely, then putting other chips in contact with it will heat them up instead of cool them.

I explained this above, so I'll just remind you that the primary mechanism of energy transfer in a heat pipe is phase change. The majority of the heat leaves through condensation, which basically puts a ceiling on the temperature to which the CPU can heat the other chips - the fluid's boiling point.

No need to be rude

I thought your post was kind of rude - to speak so negatively (and at such length) about something, on the basis of such ignorance. I guess arrogance + ignorance != rudeness, but the effect is similar.

Depending on the resin holding it in there, it might not even be more effective than direct contact to the aluminum heat sink

That's the real question, IMO. How thermally-conductive is the resin, and does the heatpipe have much direct contact with the fins? If the heat pipe has a large contact area, due to being smashed against the aluminum, then I don't really care too much about the resin. On the other hand, if the contact area is small, then the resin could be a key weakness of the design.

Heat pipes cost a lot more than the copper that they are made of. All the extra steps to inlay it into the heatsink also adds cost that wouldn't be necessary with a solid copper heatsink.

It depends on how much you're talking about, but I'll drop this point as I don't have much current knowledge on the subject.

I just know that CPU heatsinks don't use copper fins. Moreover, the last such heatsink I saw that did, had a significant price differential vs. a version of the same model that was mostly aluminum.

Edit: This video claims that a heatsink with integrated heat pipe is cheaper than equivalent copper heatsink without:

Of course, as it's by a heatpipe manufacturer, it's certainly a self-serving claim.

 

[bit_user]

most people have never gone shopping for thermally conductive tape before, and I've never found a review where they test that heatsink with better tape.

FWIW, there appears to be 4 different kinds of tape included in this review:

https://www.tomshardware.com/reviews/thermal-paste-comparison,5108-10.html

Ah, I miss Igor. He was behind some of the site's best content, but it seems they stopped using him after the last big editor/staff shakeup. I heard he still writes for the European sites.

BTW, when I mounted heatsinks on my original Pi, I discarded the included tape and used a decent thermal compound (though I forget which, exactly).

 

[bit_user]

I've designed a different solution in mid July.

Wow, that's massive!

Lack of thermal gradient at the usual load was to be expected. It shows that the heatsink itself doesn't suffer from not enough mass, but surface area.

I think he's comparing the temperature reported in the graph with the approximate temperature in the thermal image. As the heatsink is a lot cooler than the CPU, it suggests inefficient heat transfer between the two.

 

[bit_user]

Hello! Maker of the heatsink here!

Thanks for dropping by!

Please be sure to include a measurement with both CPU and GPU stress tests running.

Also, it would be nice to see how it holds up to 2.147 GHz / 750 MHz overclocking:

Raspberry Pi 4 Can Now Overclock to 2.147 GHz. Here's How. - Tom's Hardware

With a few tweaks and new firmware, the Pi 4 can get faster overclocks than ever before.

www.tomshardware.com

 

[RedShiba]

Also, it would be nice to see how it holds up to 2.147 GHz / 750 MHz overclocking:

And I thought 2ghz/600mhz was already amazing 0_0. Thanks for bringing that to my attention, I'll be sure to test with those clocks.

 

[CooliPi]

Do you guys have some good GPU test at hand? I use 4x linpack to test CPU performance. The biggest load - certainly more so than a Phoronix test suite power wise.

 

[CooliPi]

Do you guys have some good GPU test at hand? I use 4x linpack to test CPU performance. The biggest load - certainly more so than a Phoronix test suite power wise.

The whole power load of the system comprises also power consumption of the other chips. Have somebody tested stability with ethernet flooding, USB3 load (perhaps copying a large file between two fast flashdisks), wifi flood, GPU load test and 4x linpack ? That's my dream setup. I have yet to find time to test it.

 

[bit_user]

The whole power load of the system comprises also power consumption of the other chips. Have somebody tested stability with ethernet flooding, USB3 load (perhaps copying a large file between two fast flashdisks), wifi flood, GPU load test and 4x linpack ? That's my dream setup. I have yet to find time to test it.

ODROID stressed their N2 with the following commandline:

stress-ng –cpu 6 –cpu-method matrixprod && glmark2-es2-fbdev –off-screen –run-forever

Of course, you'll want to use --cpu 4.

Source: https://www.hardkernel.com/shop/odroid-n2-with-4gbyte-ram/

Beyond that, I'd use a USB3-connected SSD (not a thumb drive - SATA SSDs are much faster) and maybe just run dd to read from the raw device, in a loop. Something like:

while true; do dd if=/dev/sda of=/dev/null bs=65536; done

Put that in a script and run it with sudo.

For an Ethernet load, you could use something like netcat, though I'm not familiar with it.

 

[CooliPi]

Cool, thanks for the info about stress-ng ! I'm trying it now. Regarding glmark - it stopped compile with an error. No time to investigate now.

I also know dd and netcat, good point.

So far, with stress-ng running, the BCM temperature is 50degC, ambient 28degC. Frequency 1850MHz, over_voltage=2, one FullHD HDMI output.

I'll compare it with linpack, to test which one is a bigger power hog.