First of all, I searched community and red all posts that I found. However, some of them contain images, some of them outdated. As a blind developer, I need more and clean recommendation to continue.
Could you help me to make a selection for Mini PC to run Home Assistant?
I used Raspberry until now but need more power to use with Voice Preview/LLM.
I know that UEFI and intel NIC support is important.
Possibly I’ll install via virtualization.
I have Sonoff zigbee USB dongle and HomeAsisstant owned zigbee dongle.
I need a device to consume less power.
in our regian, we have:
•Beelink: S13, EQ14, Me mini,
•Gmktec: G2+, G3+, G5, G9
•Asus: PN43, nuc 12 slim, nuc 14 slim
You listed the only real requirement (for x86_64) in 2a, everything else is more fluid and ‘depends’ on your personal requirements; 2b is wrong (‘every’ semi-common ethernet card will do; specifically realtek ethernet cards (r8169/r8168/r8125) ones are very common and generally just work as well).
Depending on your needs, the ‘weak’ requirements are now much more important:
idle power consumption (~= the cost of running the system 24/7, 365 days a year)
peak power needs are much less relevant, important is the idle draw (something around 5 watts is possible if you choose well, 10-20 watts realistic unless the manufacturer really messed up, >60 watts easily possible if your just re-use your old gaming rig (dedicated graphics cards severely push the figures, usually +30 watts on their own)
to retain your own sanity, ~10 years old should be a hard cut-off for purchases (older gear usually has too many drawbacks, ranging from simple age/ defect rate to power/ performance ratio and support for modern standards (addon devices, replacement components, hardware accelerated features, etc.)
now voice processing drives up your requirements considerably compared to ‘just’ home-assistant (esphome builder or frigate (and many other addons) would as well, but I guess you are familiar with their needs, if you use them).
If you consider local LLMs, RAM and accelerated graphics might play a role as well (dedicated graphics and the idle power penalty coming with that)
Virtualization can help you, in the sense that you get easier access to the (virtual-) console of your home-assistant instance, but on the other hand it makes direct access to USB devices (zigbee coordinator) more difficult (it’s still possible, but not quite that easily - you might favour ethernet based zigbee coordinators instead, if you need virtualization). Another alternative to virtualization (without the drawbacks of USB device pass-through into the VM) might be having a native serial port on your x86_64, using serial console access - but this severely limits your selection (as RS232C has become rather exotic in recent decades, but some devices still have it). Another topic is your preferred access method to home-assistant, most users will probably favour the webinterface over direct hardware console (tty or serial) access. That’s something only you can answer for yourself (if you don’t need the hardware console access (regularly), the inherent advantage of virtualization to expose it to you in text form might be less relevant).
‘Normally’, virtualization only offers these features:
resource sharing, being able to run multiple independent VMs on a single system
this also comes with drawbacks, as you need to provide more RAM, CPU and SSD storage in these cases
snapshots/ backup strategies (still needs attention, doesn’t just work by itself, but makes it easier)
‘nicer’ access to the system console, should you need it instead of the webinterface (I realize this might be very relevant to you)
the drawback comes in if you need your VM to have direct hardware access (USB zigbee coordinator), most hypervisors do offer methods to pass devices through into the VM, but this does add new complications (ethernet based alternatives are usually easier in these cases).
I am fully aware that neither of this directly answers your questions, I just hope it provides some food for thought.
Disclaimer: I have played with voice (spotpear ball v1), not very successfully (as actual voice recognition is abysmal for me, but technically ‘works’ (in an unusable fashion)). I have not played with LLMs (and won’t, no I’m not getting dedicated graphics card, huge amounts of RAM/ VRAM, disk and idle power draw for this) and cannot judge about their performance requirements.
I bought Asus PN43 with 32GB RAM; N100 processor, 512GB NVME that comes with 2X2.5gbit ethernet due to low power usage.
But for Local LLM and voice processing, this device will not be enough as I imagine.
Any input would be helpful.
If you already bought it, just wait and see how it behaves in practice. From my perspective, this should be rather overpowered (at least for home assistant and voice, LLM is something I can’t answer).
I have four instances of HA, two each in two locations. One for development and one for production. All of these are running virtual machines, VirtualBox. At first blush, it does appear to be hard to pass devices through into the VM, but actually it is pretty easy.
I buy refurbished PCs (with 3yr warrantee ~USD$100) and run them headless (some need dummy VGA plug to boot) The only downside is that they run at 25W as opposed to an RPi5 at about 10W. Faster CPU and RAM, more RAM and HDD. A similar RPi5 (w/o HDD) is about USD$125
While it takes a bit of time to set them up, I’ve been running them 24/7 without any hardware problems for at least 3 years. Wait, not true, the SSD drive failed on one during the warranty period. Good thing I had a backup.
I use RealVNC to access the host OS (linux)
A laptop usually uses less power, but at twice the price
Decent mini pc but for an llm you need something like the jetson orin nano, not sure if its compatible yet but theres plenty of talk about it on the forums. It will be my last mini pc upgrade once everything is final.
Thats a very inefficient way of running Home Assistant, essentially your running a heavy operating system like windows then virtual box, it would be okay if your pc is beefy. but a more efficient way to run HA in a vm is to use proxmox. or maybe just remove the os from the old computers and install linux, then youd see a huge performance boost
Hi again,
Yes, waiting for the Asus delivery. Apart from this, I have Mac Mini M4 Pro, do you recommend using it? I believe that it will be good for local LLM.
But which is the best method to install HA on Mac Silicon?
Best.
