heat, GPIO, antenna and BT was all discussed yesterday.
I suggests you all watch the Q&A https://www.youtube.com/watch?v=KPR-shzT_VQ it will answer your questions/concerns.
1 Like
Thanks. The Q&A lacks maybe a TOC.
I don’t really need to be informed about what is a M2 slot or a CM4 anymore 
40 minute mark for thermal discussion.
1 Like
I have seen the video, and can confirm: ALL (yes, all) of the points/concerns mentioned so far, here in this thread, were all addressed in the video. They thought about those and have explained.
So, yeah, I suggest everyone go watch the video first, before calling anything a “bad design”.
3 Likes
I’ve seen the video too, and, call me a skeptic, call me a idiot, call me whatever you want, my concern is not taken away.
Future will show, how it turns out. Still, the basic though is, as the OP say’s: Passive heat-sink + absence of ventilation openings = asking for trouble.
Dozens of products in the past showed this rule is legit. I don’t know, why this box will be the first to break that rule…
The thermal throttling issues around Raspberry are widespread item, everyone does know about.
So, as usual on this forum, I need to say again, THIS IS NOT A RANT!!!, we are trying to HELP.
just my 2 cent’s… Thanks for reading this, thanks even more for THINKING about it, thanks even much more for not calling us stupid…
I watched the video and I’m still not satisfied by the answers.
What does the loadAvg in the graph really mean? Is it a good way to measure the throttling that will occur when the CPU is too hot? Why no graph with the CPU clockspeed so you can really see if it’s throttled?
And questioning if a design might be bad is not a personal attack but a warning to prevent failures, so for the greater good of Home Assistant.
3 Likes
Here is one:
2 Likes
Here’s a simple guide that is supposed to calculate the temperature rise of an enclosure based on it’s dimensions (to determine the surface area), material of construction (for its thermal conductivity), power consumption, and the external room temperature.
https://www.powerstream.com/temperature-rise-in-an-electronics-enclosure.htm
The thing is that here in NL at least alot of the times we have stone and concrete walls/ floors. So i usually install an external antenna to even get out of the server room/meter cupboard? (dont know how to translate meterkast). I’ll just stick with the nuc and stick
Well, the engineer (Stefan) comes from Switzerland. Not sure if wood is the norm, there, neither…
Whilst I agree about the heat issue, it is easily solved with some mods to the case so no big deal for me.
What I do see as a bigger deal is the NVME as it is an interesting design choice, I’m not going to say bad, just that I actively choose to not go in that direction when I was comparing all the common off the shelf ARM devices. NVME means that you have to give up USB3 to have that feature with current ARM processors. By only having USB2 on this device you now can not get decent external SSD speeds and other AI accelerators via USB3. I would personally prefer to have 4x USB3.x ports over a single NVME slot. What do others feel they prefer?
Now I know someone is going to say how much faster NVME is and that they prefer it, OK but if the data still has to pass through the 1gb network port, any gains made from NVME are lost. Interested in hearing from people that feel this is a trade off that they prefer and what they will be using it for.
4 ports of USB3 and no NVME, that is starting to sound like an Odroid N2+ only with a slower CPU 
Problem with having USB3 is that it would have interfered with the zigbee module most likely.
1 Like
Mmm… Where did you get that from?
NVME is PCIe, and hooked directly to the SoC…
Actually, besides the issue with USB3 interference, they mention in the QA that they could not have USB3 because the sole PCI lane was taken by the M.2 slot (54:10)
It’s a fact that if you have a hard requirement for an external SSD, no use for the builtin zigbee and are oblivious to the modularity of a CM4, there is little value in Amber,.
Yes but let’s talk real world end use cases and not a command line benchmark. You use the drive as a NAS = no advantage. A NVR = no advantage. If the only way to get the data on and off the drive is via a 1gb network card then How do you benefit from the extra speed when you can not connect other usb3 devices? Perhaps the next generation of ARM will allow both USB3 and NVME, we can only wait and see.
As for the CM4, you can buy a whole SBC for a similar price that has similar features. I don’t see the VALUE in it when I can have all new technology, keep the old device as a backup, and the newer SBC does not need to make compromises to try and keep backwards compatability.
Usb3 rf issues stem from bad shielding caused by buying or designing the cheapest option. There is a risk of an unknown future CMx and untested NVME drive causing similar result. If you can use a longer USB cable to move a rf dongle away from the rf noise you have an easy to implement solution.
Whilst I have not been in the industry in a long time, I was employed by a production house that did embedded designs. I can not offer much, as not much is released and the development is probably not done when you look at the shipping date of Nov 30, 2022 which is probably optimistic when you have world wide shortages going on. A Pi5 could be out before this ships.
Only measuring the CPU and enclosure temperatures is not enough, you also have localized heating of certain parts on the board. Switch mode IC, capacitors, even a resistor or PCB traces can fail. Industrial testing by a production house would include a few months of running non stop inside an oven to test and find which components get stressed and what needs to be upgraded to ensure a long life for the user. The idea is to accelerate wear and tear far above normal to find out what parts fail first. Certain countries demand 2+ year warranty by consumer laws and when you have to do rapid development to get products to the market before they are outdated and also beat competition, tests like this and more are very important to be done.
That is only 1 of many tests done. Another may be drop tests to test how the heatsink and other parts fare during shipping and also with rough handling after the end user unboxes the unit.
The released graphs show a starting temp of around 27 degrees C which is a little low if someone has it inside a badly ventilated closet or networking rack with other heat sources around. Odroid for example do 35 degree C ambient tests.
Usually the process is.
Market research to ensure people want what you are going to make. Then:
- Get a working single unit to test with.
- Small run to do stress testing on after prototype works.
- Small run to send out for beta testers and get any certifications needed, done after fixing the stress test results. If you find issues and the design needs to change, then in some cases certification needs to be redone on the updated design causing delays and big $ costs.
- Production of at least 5,000 units to get a decent price break.
- Cross your fingers there are no design flaws as you now own 5,000 units. You can do a smaller first production run to test the water and make low/no profit on the first run.
This is where crowd funding can be helpful. Market research and also gauging how many people will purchase it to plan your first production run VS the risks of getting stuck with x amount of product.
This is only the tip of the iceberg and there are many challenges. Personally I think there should be a fan header for COTS fans and the case should have slots and/or a cut away that can be punched out to mount a fan. The mold has to be made for the case anyway, its really a low cost to add as a fan header, is only a few cents, then if the user does not order with a fan, it is on them… To make up some figures out of thin air… 90% of users wont be stressing the system to the point that heat will be an issue. However when you have to offer warranty and 10% of users may need the cooling to prevent premature failures, I think the question about a fan is valid or a different way to handle it.
Who knows what the products that ships will look like exactly, so let not be rude when we really do not have all the details, don’t know what tests have been run or perhaps are being run right now. Perhaps HA will be monitoring the temp and flag a warning message to the user they need a fan, there are many ways to handle this. I wish the team best of luck and try to have fun bringing a baby to life 
I’m genuinely confused about what you have in mind.
Real-world fact: I never needed USB3 for any thing else than hooking a fast external drive to extend internal ones.
Rather than that clunky setup (like I have now,) Amber proposes to basically embed the external drive onboard via a NVME drive. It also proposes to replace my zigbee dongle by an onboard one, POE, and 32 choices regarding the SoC configuration, including BT that I need.
It just fits my needs perfectly.
Now, I’m not a fanboy.
Is it good for all purposes? Likely not.
Are there cheaper alternatives? Likely yes.
Did it make sense for NC to create an Odroid N2 clone? Definitely not.
But supporting Amber is also supporting HA, so I’m in.
What stands out to me as a point of let down is that the M.2 slot is the only option to have more storage than the otherwise tiny 32Gb, but also the best place to put a hardware accelerator card such as the Coral. With other HA server hardware devices you can have both.
This and the internal Zigbee radio. No good if you want to have the device in a server rack or similar (which I have mentioned previously).
This is not targeted at the segment of users who might have a server rack, or who are comfortable building their own Pi based server. This is a starter device to get people into the HA ecosystem for very little money, and yes it can be used for other purposes, but I would say this is the target market:
Wants to get into HA, or currently runs on a Pi, or some extra hardware they had
Wants something like a pi but with better performance and reliability
Has Zigbee devices already, or will equip their home with them or Thread/Matter devices in the future
Probably wont use an AI accelerator, the M2 slot is gonna be for fast storage
Minimal amount of hardware and software tinkering required to get it operational
Another segment of users, like me, sees this as follows:
A POE device with a Thread radio, and run Openthread and Matter stacks on the device in Docker
Can get a CM4 with no storage or wireless for it at low cost
or a wireless CM4 for BT presense detection or some other fun use
Extra USB if I want to add an RTLSDR or Z-Wave stick at some point
1 Like
…although not in that first segment I’m in the latter which is a valid user base, perhaps alongside having another HA install
Yep, I’m exactly in the latter segment as well.