Pi Zero 2W cooling tests

[g7ruh]

Hi @dylski

I have been doing some cooling tests to see how easy it is to keep the CPU from throttling.

1 is a small glued on heatsink such as: https://shop.pimoroni.com/products/heatsink?variant=17021246791

2 is a larger all round heatsink https://shop.pimoroni.com/products/aluminium-heatsink-case-for-raspberry-pi-zero?variant=40982592651347 with a heat pad linking the CPU and heatsink.

3. is a small fan designed for the Pi 4 but works well with a bit of modification (discussed below): https://shop.pimoroni.com/products/fan-shim?variant=29210095812691 Note: there is no heatsink on the CPU.

One problem is that the GPIO connector is on the upper side of the CPU and does not help convectional cooling. I have run tests with the GPIO connector at the bottom and on the side (rotating the display). Cooling is improved when no heatsink or a small one is used but not enough to stop CPU throttling.

The Fan Shim has been modified from the standard installation on a Pi 4. The fan is usually mounted on the top of the PCB. For this use case, it hits the e-ink display.

Mounting the fan on the underside of the Fan Shim PCB lowers it enough, but still leaves some space bewteen the fan and the CPU for air flow.
I have not tested changing the gap beween the fan and the CPU to see if it makes any difference to air flow / cooling: initial results were good enough.

Make sure that you do not rotate the fan when mounting. The rotating part of the fan is still uppermost. An extra header could be used instead of remounting the fan, but that increases the overall depth of the combined display and Zero 2W.

However, mounting on the underside creates a minor problem: the fan PCB touches a strengthening wing on the fan and makes the fan unable to sit square to the PCB.

Nut A clears the wing on the fan and allows the screw to be tight.
Nut B keeps the PCB above the uneven part of the fan and allows the PCB to be tightened by the bottom nut.

A suitably sized spacer could be used to replace the two nuts: those were what I had available when testing.

The PCB is thin and designed to friction fit on header block pins. I have found that is not always providing a reliable electrical connection to the PCB, especially as the outer row of contacts are elongated. The makers also suggest it should be soldered to the pins for reliability.

Care needed: do not use too much solder on the pins as that may run up the pins and prevent the header attaching reliably to the display socket.

The photo above shows the compact size of the modification. Still to be done: fitting stand-off mounts to fix the Pi to the display.

While I decide how to present the display (larger frame and find a suitable one) I used this 3D print to offer some protection to the display during testing. https://www.printables.com/model/725292-pimoroni-inky-impression-73-display-frame

Testing results:

This was tested in a room temperature of 20 - 21 deg C.
I used bcmstat.sh to monitor temperature as it provides current and maximum temperatures https://github.com/MilhouseVH/bcmstat

Test Item 1 (small heatsink) throttled the CPU ( got too hot).

Test Item 2 (big heatsink) got to 70-75 deg C and a warmer environment may cause the CPU to throttle. Not tested in higher room temperature. Using a USB attached twin fan with 3 speeds reduced the temperature but even on low speed was a bit loud. Fixing it in place would be more of a problem than 3) below.

Test Item 3 (fan) keeps the CPU to 65-66 Deg C using the automatic control and its default temperature / fan speed ranges. This is installed as a service using a supplied script after the Fan Shim software package has been installed. At low speeds the fan is not too loud.

Using a fan does add the possibility of fan failure causing problems, but using the large passive heatsink will cause some mounting issues because it is not really designed to be mounted on a display. It would require significant customisation to make a secure mount for the display.

Using the above solution and some screw headers spaced correctly will allow the display mount points to be used.

I hope the feedback helps somebody.

Regards
Roger G7RUH

[dylski]

Thanks for the detailed information! One of my setups has your the wrap-around passive heat sink (your item 2) which felt pretty safe and secure to me, i.e. it didn't clash with anything and being quite light didn't seem to add significant strain on the display via the connector. I felt the power cord was adding more strain on the than the weight of the device.

Using stand-off mounts to fix the Pi to the display would help further but I didn't feel the need at the time. If I was moving the picture frame around a lot then that's something I'd consider.

I monitored the temperature but don't recall what it got to - just that I thought it was alright, so probably around the 75 degree max you saw.

Thanks again for your research!

[g7ruh]

@dylski

Thanks for your reply.

I felt that unsecured Pi Zero "hanging off the back" would come off over time if not secured, just from heat / cold cycles pushing it off. I looked at length at the solid case and decided it would be quite hard to fix as the top half where the fixing screws enter is very thin and may not stand widening to 2.5 mm threads (they are 2mm fixings). Spending a lot of time in the workshop was the alternative if the Fan Shim did not work.

The only thing I have found is that it may (not tested) matter which order the PaperPiAI and Fan Shim software are installed. I am not sure as it only happend once and re-installing Fan Shim software restored the fan operation and the picture generating and display still work fine.

I have just completed the stand-off mounting. Using one of the supplied (short thread) stand-offs for the non-Shim end is correct. Using a shorter stand-off with a longer thread fits between the Pi Zero and the Shim PCB. the long thread can only go a short way in because the Shim PCB stops it. Doing it up a little makes it hold in the display but not too far in. If it was tight against the Shim PCB it would break it.

I am happy now because it is held in place and the other end is tight.

Glad to share and help, my way of adding something to your hard work. If you want to mention the closed issue or point to it in the Read.Me as a way to help other folks, please do. After all it is more helpful information than an issue per se.

Regards
Roger G7RUH

[Delph]

I recently filed an issue on the OnnxStream repository which is somewhat related, as I've found my WiFi connection getting dropped and not resolving afterwards.

I bought one of these heat sinks (Amazon), and a header extender which lifts the e-paper HAT well off the Pi. I haven't really had it set up long enough for any results yet.

Thanks for the good research.

[g7ruh]

I've found my WiFi connection getting dropped

Hi @Delph

Thanks for the reply, happy to help share information, it makes the world go round after all!

Interesting design of the heatsink you selected since it is top side only whereas mine has metal all round.

I think part of the WiFi dropping is signal strength related. While I was testing the PaperPiAI software, I found that if I got up and stood over the 'Pi it sometimes hung up. (WiFi dropped, did not recover, did not respond to pings from another device).

I have seen this before.

Thinking of the equipment and component design, there is a lot of metal in the display and if the WiFi source has the display between it and the Pi Zero 2W it will attenuate the signal. Add to that the 'Pi antenna is a very simple (but very clever) design as it sits on the PCB with no copper etching over it. Look at the base of the PI Zero 2W and follow the Raspberry Pi logo to its bottom, you can see a small triangle with no copper present. On the top of the PCB there are 4 small surface mount components. That is where the antenna is located. If it is covered by metal (the display, heatsink etc) the signal is going to be attenuated to some extent. Apologies if you already know antenna theory.

I have had problems with a Pi Zero W (not the 2W) dropping out. I replaced it (disabled onboard WiFi) and connected a WiFi dongle to the USB port via a short USB extension cable, no more connectivity problems on that device. It was located in a weak WiFi signal area.

I added a plastic shim to the bottom of the Pi heatsink I was using ( big one in the picture) and it seemed to make a difference: allowing more signal in and out: larger air space between antenna and metal heatsink.

I know this is all subjective, but as a radio amateur I do understand radio frequency theory somewhat. Add to that the Pi Zero 2W only supports 2.4 Ghz WiFi and why do microwave ovens use 2.4 GHz? because water gets hotter at that frequency and so if a human gets in the way of the WiFi signal it will be absorbed too.... humans are mostly water after all.

I think we need to consider signal strength and WiFi dropouts being part of the same kind of issue, causing drop-outs and connectivity hangs.

When I had this issue during testing with the test items in the photo, I also used a micro USB to USB A adapter that has an ethernet socket too. That way I was able to negate WiFi issues during initial testing.

Now I am using the Fan Shim, there is lots of air space around the antenna to let the WiFi signal in and out and less metal around it all. I have not had any hangs or ssh sessions to a remote session stopping.

I got to the stage with the big metal heat sink that I was using during testing that it was hanging and I was not sure if it was software or WiFi signal related issues. So I was using /home/rp/PaperPiAI/venv/bin/python PaperPiAI/src/generate_picture.py ..... with "&" on the end so that if the ssh session dropped, it may at least finish the picture generation without dropping the session and killing the generate job. That allowed me to continue heat related tests.

Hope that helps

Regards
Roger G7RUH

[Delph]

So, in my setup, the Pi is on the back of a photo frame (I haven't used the Inky Impressions display, just a Waveshare one) and exposed to air. It's about 1-2m from my access point, LOS. I only added the heatsink yesterday as I wanted to rule out it being a throttling issue causing the WiFi driver to miss kernel messages (as you'll see in the linked issue), and then get into a non-recoverable state (even 12 hours later, there's still nothing, but a reboot and it is fine).

I'll need to leave it a few days and see if it still occurs, but I had also collected some data on cpu temp without the heat sink, so when I grab some with the heatsink I'll report back here too.

[g7ruh]

@Delph

I notice from the picture of the heatsink you have bought, that it has a better design WiFi RF-wise. The metal is clear of the WiFi antenna.

I await your findings.

Roger G7RUH

[g7ruh]

I mentioned in an earlier post that there may be an issue with the Fan Shim not working correctly.

There is a conflict between Fan Shim and the picture display script. Both the Fan Shim and the Inky screen use GPIO 17 (Pin 11).

After running a display picture, the Fan Shim stops working.

A work-around is to restart the Fan Shim service:

# sudo systemctl restart pimoroni-fanshim.service

after the display script has finished.

Display Pinout

Fan Shim Pinout

From https://pinout.xyz/pinout/inky_impression and https://pinout.xyz/pinout/fan_shim

[Delph]

I haven't captured any temperature data yet, but I'm still experiencing the network failures. I haven't had much time of late.