dan s
Posts: 132
Joined: Tue Feb 18, 2020 10:42 pm

opinions on cm4 heatsink design/requirements

Tue Sep 14, 2021 3:58 am

I picked up a cm4 and a bunch of paraphernalia a little while back for a long term project.

For cooling I went with a 30mm pwm fan, the gearwork heatsink, and a custom python script to control the fan.

I've got everything working pretty well, even in a very small restricted enclosure (see below)
stress.png
stress.png (49.82 KiB) Viewed 1252 times
With that being said, I'm not completely happy with the heatsink. I feel it's a little bulky and acts more like a thermal reservoir than heatsink. Not to mention the fin orientation is less than ideal for my enclosure.

I started researching what others have done, and people are cooling all kinds of stuff.
  • SOC - as you would expect
  • RAM - also expected
  • EMMC
  • Ethernet controller
  • PMIC
Thus my question is what (in your opinion) really needs cooling, and what's just overzealousness/marketing?

pidd
Posts: 2636
Joined: Fri May 29, 2020 8:29 pm
Location: Wirral, UK
Contact: Website

Re: opinions on cm4 heatsink design/requirements

Tue Sep 14, 2021 4:34 am

In the example graph, it is cooling rather than acting as a reservoir for the majority of the time. If it were acting as a reservoir there would be a continual upwards slope in either or both of the PWM and temperature whereas it is showing a horizontal graph for both the majority of time indicating a permanent thermal equilibrium has been achieved.

aBUGSworstnightmare
Posts: 3635
Joined: Tue Jun 30, 2015 1:35 pm

Re: opinions on cm4 heatsink design/requirements

Tue Sep 14, 2021 5:31 am

Only the SOC requures cooling.
I'm using the Geekworm heatsink wirh 40mm fans, see here i.e. viewtopic.php?f=29&t=313280&p=1880561&h ... k#p1880979 . Yes, I have no size constraints. If you're able to make your own heatsink, possible as one can see here viewtopic.php?f=98&t=311607&p=1863681&h ... k#p1863969
Make sure your fan is sucking air away from the module and not blowing into/on the sink.

Or, try a different solution, waveshare offers two https://www.waveshare.com/product/raspb ... atsink.htm

dan s
Posts: 132
Joined: Tue Feb 18, 2020 10:42 pm

Re: opinions on cm4 heatsink design/requirements

Tue Sep 14, 2021 12:52 pm

pidd wrote:
Tue Sep 14, 2021 4:34 am
In the example graph, it is cooling rather than acting as a reservoir for the majority of the time. If it were acting as a reservoir there would be a continual upwards slope in either or both of the PWM and temperature whereas it is showing a horizontal graph for both the majority of time indicating a permanent thermal equilibrium has been achieved.
I wasn't meaning a reservoir in the strictest sense, hence why i said acts like.

The reason why I mentioned reservoir, is because it takes what I would consider an exceptionally long time for the SOC to come back down to pre test temperatures. In the graph above The idle time pre and post test was 30 minutes. It took the full 30 minutes for the SOC to come down to it's pre test temperature, and i think that's to long. Even if blip the fan to 100% right after the test finished it will still take 10 minutes or so to bring the temp back down. That is to long in my opinion and i think it is the nature of the heatsink thats is causing it.

The heatsink is designed to mount a 30mm fan , so thats what im using. However as designed, the fan can't pull air through all the fins. It can only pull air through roughly 60% of the fins. That's why I think you see slow ramp up and ramp down sections of the graph. Heat is flowing into the 40% of the sink that can't be cooled via the fan and acting as a reservoir. After the test the fan has to work to pull the heat out of those sections.

dan s
Posts: 132
Joined: Tue Feb 18, 2020 10:42 pm

Re: opinions on cm4 heatsink design/requirements

Tue Sep 14, 2021 12:56 pm

aBUGSworstnightmare wrote:
Tue Sep 14, 2021 5:31 am
Or, try a different solution, waveshare offers two https://www.waveshare.com/product/raspb ... atsink.htm
I'm using a waveshare Mini Base Boad (b version), and it's funny you mentioned them.

For both there cooling solutions they show sinking the PIMC. ON their fan they show sinking the SOC, RAM, PIMC, and ethernet controller.

They are actually one of the reason i went down the rabbit hole of what actually need cooling.

Image

Image

aBUGSworstnightmare
Posts: 3635
Joined: Tue Jun 30, 2015 1:35 pm

Re: opinions on cm4 heatsink design/requirements

Tue Sep 14, 2021 1:56 pm

The 30mm fan is to small to be effective.
Image

As you can see on the picture I have the 40mm fan mounted on 5mm stand-offs to allow for convection in addition.
With my 40x40x20mm fan it takes just moments for the SOC to come back to lower temps. Running an overclocked CM4 (2GHz) on a full kernel compile results in max. 45°C SOC temp (tray icon readings).
That fan pulls roughly 10m3/h air.

dan s
Posts: 132
Joined: Tue Feb 18, 2020 10:42 pm

Re: opinions on cm4 heatsink design/requirements

Tue Sep 14, 2021 3:49 pm

aBUGSworstnightmare wrote:
Tue Sep 14, 2021 1:56 pm
The 30mm fan is to small to be effective.
Hardware wise im limited to 5V pwm fans, as that's all the waveshare I/0 boards supports. Design wise I'd like to keep the enclosure as compact as possible. I have the tools at my disposal to design and manufacture just about anything. Given what's going on in the world and the state of various supply chains it's pretty hard to get you hands on small 5v pwm fans.

Though I hate the color, I might have to go with a larger Noctua 5V pwm fan and case mount it.

aBUGSworstnightmare
Posts: 3635
Joined: Tue Jun 30, 2015 1:35 pm

Re: opinions on cm4 heatsink design/requirements

Tue Sep 14, 2021 6:49 pm

SUNON MF40100V21000UA99, 40x40x10 mm, 5 V-
https://www.pollin.de/p/axialluefter-su ... 5-v-320693

Should give you somehow equal performance.

BeauSlim
Posts: 144
Joined: Mon Jul 31, 2017 10:02 am

Re: opinions on cm4 heatsink design/requirements

Sun Sep 19, 2021 1:40 am

Those really flat, thick heat sinks are designed for bursty loads and slow dissipation with low noise. Not great for fast dissipation under constant load. And they trap air underneath them.

I like to use recycled "Northbridge" sinks from old PCs. They are usually 38x38mmx30mm (or so). They have tall, thin fins and often have mounts for 40mm fans. New ones can be found on AliExpress for not a lot.

Use the thinnest thermal pads you can, if not just thermal tape or paste.

As for fan direction, I generally find that blowing onto the sink is better, but test, test, test.

dan s
Posts: 132
Joined: Tue Feb 18, 2020 10:42 pm

Re: opinions on cm4 heatsink design/requirements

Sun Sep 19, 2021 3:23 pm

BeauSlim wrote:
Sun Sep 19, 2021 1:40 am
Those really flat, thick heat sinks are designed for bursty loads and slow dissipation with low noise. Not great for fast dissipation under constant load. And they trap air underneath them.
From the research I've done a pin style heatsink might be better as it pulls air from any direction. I've designed a pin heatsink, now i just need to get the time ro runs a thermal simulation or 2 on it.
1.png
1.png (168.17 KiB) Viewed 957 times
2.png
2.png (177.41 KiB) Viewed 957 times

User avatar
MikeDB
Posts: 584
Joined: Sun Oct 12, 2014 8:27 am
Contact: Website

Re: opinions on cm4 heatsink design/requirements

Sun Sep 19, 2021 4:00 pm

I've found with a 40mm 5V fan placed about a centimetre above a CM4 running DSP processing with constant memory and I/O accesses, there's still no need for any heatsink metalwork to keep the thing cool enough.
Always interested in innovative audio startups needing help and investment. Find me on our website.

dan s
Posts: 132
Joined: Tue Feb 18, 2020 10:42 pm

Re: opinions on cm4 heatsink design/requirements

Sun Sep 19, 2021 4:24 pm

MikeDB wrote:
Sun Sep 19, 2021 4:00 pm
I've found with a 40mm 5V fan placed about a centimetre above a CM4 running DSP processing with constant memory and I/O accesses, there's still no need for any heatsink metalwork to keep the thing cool enough.
Are you talking about in an enclosure, or in open air?

aBUGSworstnightmare
Posts: 3635
Joined: Tue Jun 30, 2015 1:35 pm

Re: opinions on cm4 heatsink design/requirements

Sun Sep 19, 2021 4:50 pm

dan s wrote:
Sun Sep 19, 2021 3:23 pm
BeauSlim wrote:
Sun Sep 19, 2021 1:40 am
Those really flat, thick heat sinks are designed for bursty loads and slow dissipation with low noise. Not great for fast dissipation under constant load. And they trap air underneath them.
From the research I've done a pin style heatsink might be better as it pulls air from any direction. I've designed a pin heatsink, now i just need to get the time ro runs a thermal simulation or 2 on it.

1.png
2.png
you should compare it against a design which has solid fins. Would be interrested to see which one performs better.

Idea is nice, but how do you want to get it produced? Milling it will take huge time, diecast might be expensive if you plan small volumes.what is the red part used for?

dan s
Posts: 132
Joined: Tue Feb 18, 2020 10:42 pm

Re: opinions on cm4 heatsink design/requirements

Sun Sep 19, 2021 5:25 pm

aBUGSworstnightmare wrote:
Sun Sep 19, 2021 4:50 pm
you should compare it against a design which has solid fins. Would be interrested to see which one performs better.
I'd got 3 models to test right now.

1) the geekworm one (found a model online)
2) one with simple fins running 90 degrees to that of the geekworm one.
3) the pin fin version

Something else I've read that might help is shrouding the fan/heatsink. By that i mean putting something over the top of the heatsink so that air has to pass through all fins to exit the case via the fan.
aBUGSworstnightmare wrote:
Sun Sep 19, 2021 4:50 pm
Idea is nice, but how do you want to get it produced? Milling it will take huge time, diecast might be expensive if you plan small volumes.what is the red part used for?
I have access to a a very robust cnc router, and it wouldn't take to long on that because of its high speed and feed capabilities.

User avatar
MikeDB
Posts: 584
Joined: Sun Oct 12, 2014 8:27 am
Contact: Website

Re: opinions on cm4 heatsink design/requirements

Sun Sep 19, 2021 10:33 pm

dan s wrote:
Sun Sep 19, 2021 4:24 pm
MikeDB wrote:
Sun Sep 19, 2021 4:00 pm
I've found with a 40mm 5V fan placed about a centimetre above a CM4 running DSP processing with constant memory and I/O accesses, there's still no need for any heatsink metalwork to keep the thing cool enough.
Are you talking about in an enclosure, or in open air?
In a fairly large die-cast box. The fan spreads the heat quite evenly around the box.
Always interested in innovative audio startups needing help and investment. Find me on our website.

BeauSlim
Posts: 144
Joined: Mon Jul 31, 2017 10:02 am

Re: opinions on cm4 heatsink design/requirements

Mon Sep 20, 2021 8:50 pm

dan s wrote:
Sun Sep 19, 2021 5:25 pm
Something else I've read that might help is shrouding the fan/heatsink. By that i mean putting something over the top of the heatsink so that air has to pass through all fins to exit the case via the fan.
Laptops pretty much do it that way. Shrouds definitely help, but I suspect that on smaller scales you run into more air friction/restriction, which is why you see a lot of blower/snail fans to provide the pressure necessary to get air moving.

I am also designing a case for the Waveshare board. I want it as compact as possible while cooling both the CM4 and an NVME SSD in the m.2 slot. I am using a 30mm fan mounted at the back (on the usb-c end) that blows horizontally over and under the board, and therefore over both heat sinks. The sides, top, and bottom are completely sealed via a vase-mode 3d-printed tube, and all air exits via vents at the ports end.
Obviously this makes external access to gpio pins and camera headers a problem, but that's fine for my use-case.

dan s
Posts: 132
Joined: Tue Feb 18, 2020 10:42 pm

Re: opinions on cm4 heatsink design/requirements

Tue Sep 21, 2021 2:59 am

BeauSlim wrote:
Mon Sep 20, 2021 8:50 pm
Laptops pretty much do it that way. Shrouds definitely help, but I suspect that on smaller scales you run into more air friction/restriction, which is why you see a lot of blower/snail fans to provide the pressure necessary to get air moving.
There are some nice 40x40x10 5v pwm fans on the market with a good volumetric flow and static pressure, they're just all out of stock because of recent events.


BeauSlim wrote:
Mon Sep 20, 2021 8:50 pm
I am also designing a case for the Waveshare board. I want it as compact as possible while cooling both the CM4 and an NVME SSD in the m.2 slot. I am using a 30mm fan mounted at the back (on the usb-c end) that blows horizontally over and under the board, and therefore over both heat sinks. The sides, top, and bottom are completely sealed via a vase-mode 3d-printed tube, and all air exits via vents at the ports end.
Obviously this makes external access to gpio pins and camera headers a problem, but that's fine for my use-case.
I'm doing the same. I'm working on a little desktop server, and a dual HQ camera version. The renders I posted above where just a fist pass at an open air test bench.


Here are some shots of the fully enclosed desktop case i designed and used for the thermal benchmarks. its 5.1 mm longer than the board itself, and 21.1mm wider. The extra width is because it has 8mm of free space on both sides. Air enters the case through the inlet on the bottom (directly over the m.2), and then comes up around the sides, through the heatsink, and then vents out the top via the fan. I'm sure I could get some passive cooling benefits if I add vents to the top.

The feet are designed to hide the screws that hold the case together. They also allow you to press rubber balls in on top of the screws to act as vibration/noise dampeners.

Images are liked to high res versions on imgbb.
Image Image Image

dan s
Posts: 132
Joined: Tue Feb 18, 2020 10:42 pm

Re: opinions on cm4 heatsink design/requirements

Tue Sep 21, 2021 3:01 am

This is the camera case version, The top is the same as the desktop version, the back is just different as it needed to flare out so that I could properly mount 2 HQ cameras.

Image Image

BeauSlim
Posts: 144
Joined: Mon Jul 31, 2017 10:02 am

Re: opinions on cm4 heatsink design/requirements

Thu Sep 23, 2021 9:38 pm

Looks good. I like that the flare on your camera version looks like it also acts as a stand to angle the cameras a bit.

I can't share mine yet. Printer hot-end went kaboom on me. Replacement should be here Monday.

Speaking of component shortages, the 2242 SSD I ordered still hasn't arrived, so I'm testing with an 80 mm desktop one, and I find that it gets quite warm. How do you find temperatures on your Samsung SSD under load?

StonedEdge
Posts: 83
Joined: Wed Oct 28, 2020 11:42 am

Re: opinions on cm4 heatsink design/requirements

Fri Sep 24, 2021 1:45 am

Following on from this conversation regarding heatsinks, this looks pretty interesting.

He claims temps of 45c under high loads, and there is no fan or heatsink setup. Only a thermal pad and thermal paste connecting to what I presume is 2oz copper ground plane on the PCB.

I wonder if those temperatures are reasonable or not. Seems to good to be true if you ask me.

https://www.youtube.com/watch?v=p_hAlw-e8WQ&t=19s

aBUGSworstnightmare
Posts: 3635
Joined: Tue Jun 30, 2015 1:35 pm

Re: opinions on cm4 heatsink design/requirements

Fri Sep 24, 2021 6:09 am

StonedEdge wrote:
Fri Sep 24, 2021 1:45 am
He claims temps of 45c under high loads, and there is no fan or heatsink setup. Only a thermal pad and thermal paste connecting to what I presume is 2oz copper ground plane on the PCB.

I wonder if those temperatures are reasonable or not. Seems to good to be true if you ask me.
can't believe that these temp readings are for real. At least it's not in line with what I'm seeing for a CM4 in free air (sitting on a CM4IO).

Question though is: what is high load? To me it's i.e. A full kernel compile as this uses 100% of the CPU for roughly 2hours (on a CM4). If you don't have cooling you will see throttling quite soon --> compilation time increases.

dan s
Posts: 132
Joined: Tue Feb 18, 2020 10:42 pm

Re: opinions on cm4 heatsink design/requirements

Fri Sep 24, 2021 6:58 am

BeauSlim wrote:
Thu Sep 23, 2021 9:38 pm
How do you find temperatures on your Samsung SSD under load?
I've only tested in my little test bench style set-up, and thats means the m.2 is only being cooled by convection.
Image


To test it I used, Jeff Geerling's benchmarking script.
https://github.com/geerlingguy/raspberr ... nchmark.sh

For the random reads/writes it never goes above ambient 19-21°C

For sequential reads, If i pound on it for several minutes it will climb into the 60-61°C range and then levels out. The max operating temp is 70°C, so I should be fine. Temps should be even better when it's in the enclosure with air actively being pulled across it.

dan s
Posts: 132
Joined: Tue Feb 18, 2020 10:42 pm

Re: opinions on cm4 heatsink design/requirements

Fri Sep 24, 2021 7:08 am

aBUGSworstnightmare wrote:
Fri Sep 24, 2021 6:09 am
StonedEdge wrote:
Fri Sep 24, 2021 1:45 am
He claims temps of 45c under high loads, and there is no fan or heatsink setup. Only a thermal pad and thermal paste connecting to what I presume is 2oz copper ground plane on the PCB.

I wonder if those temperatures are reasonable or not. Seems to good to be true if you ask me.
can't believe that these temp readings are for real. At least it's not in line with what I'm seeing for a CM4 in free air (sitting on a CM4IO).

Question though is: what is high load? To me it's i.e. A full kernel compile as this uses 100% of the CPU for roughly 2hours (on a CM4). If you don't have cooling you will see throttling quite soon --> compilation time increases.
yea, I would question those temps as well.

A lot of people run stressberry with the default settings and look at the maximum temperature reached. When you look at the graphs you can see they aren't anywhere close to equilibrium.

BeauSlim
Posts: 144
Joined: Mon Jul 31, 2017 10:02 am

Re: opinions on cm4 heatsink design/requirements

Sat Sep 25, 2021 7:11 am

The game machine doesn't seem completely unreasonable. Many of those game emulators aren't that high load and run reasonably well even on a Pi Zero if the display is low-res. In the video's comments he suggests that they are taking temperature into account, going so far as to move WiFi to an external card to reduce heat from the CM4.

Worst case they can throttle performance like Apple did in some of their late Intel laptops. Or machine the case out of aluminum and charge an extra $100. ;-)

Return to “Compute Module”