Nabu Casa’s primarly income is the Home Assistant Cloud service subscription, not the hardware sales, so they would in theory make more money in the long term if they practically gave away a entry-level (low-end) model at cost, meaning that sell it for what it cost to manufacture it without making a profit on that hardware. In practice Nabu Casa as a company will make more money as the Home Assistant userbase grows as long as there are then also more people who subscribe to the Home Assistant Cloud service subscription. → Home Assistant Cloud - Home Assistant
That business model is referred to as the “Razor and blades model” and is today also the same concept commenly often partially used for video game console sales, where they make very little money on the hardware and instead make money on → Razor and blades model - Wikipedia but for video games they also make a lot of money on both games (software) and other attachments such as game controller → Attach rate - Wikipedia
So in the case of Nabu Casa there will probably a larger profit margine in not only the Home Assistant Cloud service subscription but the sales of official Home Assistant branded hardware such as example existing IoT radio USB dongles (for Zigbee, Z-Wave, and Thread) as well as the upcoming smart speakers / voice assistant appliances (and maybe on-premise AI-accelerator solutions) in the future. There is also Economies_of_scale to consider if they for example make and sell a lot more smart speakers than they sell boxes that run Home Assistant → Economies of scale - Wikipedia
Anyway, from a business point-of-view I suspect that it should be much smarter to have a higher profit margin on mid-range and high-end hardware models than a low-end hardware model because the more expensive something is and the fewer total number of each product you manufacture of it he larger profit margin on each such product sold you can have with people still buying it → Profit margin - Wikipedia
Hi @agners , hope you can help - I just got a brand new HA Yellow with CM5 and I’m actually in doubt where to position my thermal pads since the guides only show CM4 - could you assist if this is the right position of thermal pads?
Best, Martin
Because of the height, you have to put the quadratic one onto the BCM2712 SoC (silver), and the square (with the longer edges) onto the memory chip (in your picture just above the BCM2712 SoC, with the M marked.
For users who update: Your thermal pads might be dried out or just broken otherwise. I’ve run some tests with and without thermal pads and CM5, in my testing it seems that makes a very minimal difference. This are tests under full load:
It seems that the rather large and flat aluminum surface of the BCM2712 provides quite good heat transfer on its own. So operating without thermal pads should be totally fine too.
@agners do you think that you at Nabu Casa could make and sell such dual M.2 adapter (with a PCIe switch IC) for the Home Assistant Yellow so we can use both an NVMe SSD and an AI accererator card at the same time?
I have not seen anyone else who sells exactly such an “Single slot to dual M.2 adapter” adapter but found this “NVME to dual SSD adapter” which does not have the same functionality since it is only converts one NVMe slot to two SATA module slots but think it might still be good as a visual reference of a stacked form-factor concept:
@agners If you look at the pictures of it then I think you will at least understand the concept where you would instead have new dual adapter that allows the user to add NVMe SSD underneath in the stack and the AI accelerator stacked on top of it via the dual M.2 slots that would need to be stacked with an off-set for that.
Again, there are otherwise plenty of other single-to-dual adapter products (or even single-to-three or single-to-four) with PCIe switches for the Raspberry Pi that split one M.2 slot into two, but none that I found have this type of stacked form factor which would allow us to fit two cards in the Home Assistant Yellow-.
Thanks a lot - that makes perfect sense especially taking the thickness of each pad into account (I wondered that when I looked at the thermal pads).
Most appreciated!
New user here. Just got the home assistant yellow with power supply, eu, and a cm5 raspberry pi. Installed it using rpboot and then installed the software on the device (using option 2). However connecting it to Ethernet and power supply I just get a short flash of red light and then a solid green light and nothing happens. Do not see it on the deco tp link I got so no IP? What to do?
I am in a similar situation. I am replacing my CM4 with 32GB of EMMC plus NVME with the CM5 4G RAM with 32GB EMMC and Wi-Fi plus NVME and when I follow the instructions all I get is a power on red led once then goes solid green.
Maybe?: In Raspberry Pi Imager (1.8.5) I need to choose “no filtering” to find the “Home Assistant OS Installer for Yellow” image, however, it was last updated 10\2023 and specifically for the CM4. Is there another location to grab the latest version? Are there steps we are missing or not quite understanding?
I know this is still bleeding edge and it will become easier in the future but, hoping there may be something we are missing or misunderstanding.
Here is an add to those wondering about using the official CM5 heatsink.
The heatsink purchased separately for the CM5 works great in the Yellow. To install it, first make sure the CM5 is not screwed in then just align the heatsink with the notch towards the edge the the Yellow PCB and use the long screws provided with the new heatsink. Screw in from the bottom of the yellow making sure to add a little pressure to the heatsink and yellow (sandwiching the CM5) so it is fully flat to the CM5. This is because the holes and heatsink are both threaded so the screws will not help pull the heatsink. The screws do not interfere with putting on the case base plate
I had the same issue with mine and had to resolve to the “backup”-guide where you download the latest stable img from the git repo yourself and use that instead of the one from the Raspberry Pi Imager - and I couldn’t find any note on it anywhere what “just a red LED” meant.
So I had the HA directly cabled via USB-C and moved the jumper.
When I used that it worked afterwards.
Perfect That did the trick. I was on the right track of needing a newer version of the Yellow image but I failed to find it when searching as I did not even know if there was one. Maybe the OP can update the “Installing on CM%” to point to the latest github release.
I think there may still be issues with the CM5 with eMMC and Yellow hardware.
I could not get it to boot from the onboard eMMC at all, when it was tried on a carrier board with a HDMI port, it seemed to hang waiting for the base filesystem to become mountable, I left it quite a while but it never mounted - the partuuid was present, it’s like the kernel couldn’t even see the eMMC.
I could only get it to boot by doing rpiboot writing the HAOS image to the NVMe directly.
Do make sure you get the HAOS 14.0 Yellow image from GitHub releases though, the one that Raspberry Pi Imager provides does not work with the GPIOs, otherwise you’ll get issues like the one below.
When it did boot, the status LEDs seems to only have a solid green working, neither of the other two did anything, HA was running when I went to the IP address. Unfortunately though, it seems to have lost the /dev/ttyAMA1 entry for the onboard Zigbee radio, only /dev/ttyAMA10 showed up, which I think is the UART, as ZHA could not use it, it just refused to bind. On the CM4 there is ttyAMA0, ttyAMA1 and ttyAMA2
So while it may boot with a CM5 Lite and seem to work in the most part (and quick too), I think there may be some subtle differences with the CM5 with eMMC (all I could get in the UK) which means some things may not quite work as expected, so I’ll keep an eye out for future updates before I try again
Luckily I used the separate hardware, so I just reverted back the CM4 for now.
The image which was available on Raspberry Pi Imager so far was mainly meant to be flashed on a USB flash drive and only contains an installer. This installer is not compatible with the CM5 on Yellow.
Since today the Raspberry Pi Imager offers the HAOS image which can be directly flashed onto the CM5 eMMC (or NVMe for the light variant) when using rpiboot:
I think it would be beneficial for all Yellow users if the next revision for CM5 Modules will accommodate holes pattern for standard CM5 Cooling fan from Raspberry itself!
I had assumed that “yellow support for cm5” also meant “generic cm5” support but that does not seem to be the case.
Neither the generic raspberry release nor the yellow specific one work on a cm5 (using eMMC) when using a Compute Module 5 IO Board instead of a yellow.
I’m having network issues with my current setup and I bought the CM5 Dev kit for HAOS. I hope there will be a working release for the “generic CM5”, since I’ve spent too much money for the dev kit to just replace it with Yellow.
So there are issues using the new Home Assistant OS 14 on Raspberry Pi’s official Compute Module 5 Development Kit with their Raspberry Pi Compute Module 5 IO board?
Well at least with my limited knowledge I couldn’t get it to boot. I tried both eMMC and NVME and both images on Pi imager. The image for the Yellow just loop restart w/ u-boot icon appearing on the screen and the RPI 5 image does load somewhat, but it halts when there is a High Speed DDR5 recognize or something.
Does that mean you are using RPIBOOT method described in the “Option 2: Reinstall Home Assistant OS using rpiboot” + “Using Windows Installer…” there? → How to reinstall the operating system – Home Assistant Yellow and if so have you also tried to the even more complicated “From source code…” way too?
I haven’t spent the time to troubleshoot but I’m going from a CM4 (4GB RAM, No eMMC, with WiFi, booting from M.2 SSD) to a CM5 (4GB RAM, 32 GB eMMC, WiFi, boot from M.2 SSD) and sadly my HA Yellow will not boot. Looks like I may need to re-image and restore from backup like others have done.
That did the trick. I was on the right track of needing a newer version of the Yellow image but I failed to find it when searching as I did not even know if there was one. Maybe the OP can update the “Installing on CM%” to point to the latest github release.