Replies: 1 comment
-
|
By the way, while on the topic of the Home Assistant Yellow and M.2 AI accelerators, does anyone know if there a compatible dual M.2 PCIe switch adapter that you can buy for the Home Assistant Yellow (and other CM4/CM5 carrier boards with only a single M.2 slot) that would allow us to use both a Hailo-8 AI M.2 module and a NVMe SSD harddrive in it at the same time? The same use case would also suit the Raspberry Pi Compute Module 5 IO Board (and the Raspberry Pi Compute Module 5 Development Kit which comes with it) as it too have no way to add both a AI accelerator M.2 module and a M.2 SSD harddrive at the same time The point with that is it would be awesome for Home Asssistant Yellow users (and users of the Raspberry Pi Compute Module 5 IO Board) could use a M.2 AI accelerator card it without sacrificing the NVMe SSD storage M.2 slot, thus the need for some kind of PCIe switch adapter in a nice slim format. There are at least some dual M.2 HAT(s)/Shield(s) with PCIe switch chip for the regular Raspberry Pi 5 that allow you to use a Hailo-8 AI accelerator (Raspberry Pi AI Kit) and a M.2 NVMe SSD at the same time? See example Geekworm X1005 PCIe to M.2 NVMe SSD PIP Bottom for Raspberry Pi 5 which by reports looks to also support Hailo-8 AI Accelerator but that adapter does not have an M.2 input interface: Perhaps Nabu Casa could make and sell such dual M.2 adapter with an onboard 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 at least 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: If you look at the pictures of those bellow then I think you will at least get an idea of the form-factor 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-. |
Beta Was this translation helpful? Give feedback.


Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
-
Anyone tested the Waveshare Hailo-8 M.2 AI Accelerator Module?
Will the Waveshare Hailo-8 M.2 AI Accelerator Module for Raspberry Pi 5 compatibility with the Home Assistant Yellow hardware (and Home Assistant Operating System) out-of-the-box? And even if it will work is there a risk it will get too warm/hot inside the Home Assistant Yellow case?
For reference, that in turn is Waveshare variant of reference Hailo-8 M.2 AI Accelerator which you can find in Raspberry Pi PCIe Database:
Backstory is that it was previously announced that the Home Assistant Operating System 14 brought support for the Hailo-8 AI accelerator, and at that time the only option available on the market was the M.2 AI accelerator card found in the Raspberry Pi AI Kit, but that only features the low-end "Hailo-8L" chip with just 13 TOPS computing power variant in the Hailo-8 AI accelerator chip series, which is a lot less compared to the faster Hailo-8 AI accelerator chip used in the Raspberry Pi AI HAT+ which gives at up to 26 TOPS compute power, but now Waveshare have released their Waveshare Hailo-8 M.2 AI Accelerator Module with the same chip which at least looks to have the correct form-factor, though the question is if it in practice will work with the Home Assistant Yellow enclosure and the latest HAOS?
The specification sais Power consumption is typically 2.5W but that it can draw up to 8.65W (Max).
If you have a normal Raspberry 5 then they also have Waveshare PCIE M.2 HAT+ for which they posted instruction on getting running:
https://www.waveshare.com/wiki/Hailo-8
https://github.com/hailo-ai/hailo-rpi5-examples
Waveshare wiki looks to indicate that it uses the same PCI identifiers as the AI KIT+ (
1e60:2864), and the guide doesn't mention any specific driver needed on RPi OS. It calls fordtparam=pciex1_gen=3though I read a report of someone running the Raspberry Pi AI Kit without that. So I would assume that from the software perspective this card should be supported out-of-the-box in latest HAOS versions?PS: I know that this is much more expensive than the Raspberry Pi AI HAT+ which uses the same chip, (alsoa way more expensive solution than PC-based solution with a GPU), but if you want to use Home Assistant Yellow in an edge-installation or a showcase demo and POC unit they it might still be worth it. And it would of course also be more cost effective to just buy a Raspberry Pi 5 and their Raspberry Pi AI HAT+ for the same use case. So please respect that the question on wheather or not is whether or it is worth it is not really the question up for discussion here.
Beta Was this translation helpful? Give feedback.
All reactions