I like the zero-issue option. Currently, my system is not so widely spread that I cannot control something without HA integration, but as soon as I add window blinds and the boiler to the setup stability will be paramount.
@k8gg I want to learn Proxmox, Traefik, and some other tools too. Thank you for the valuable resources, but I will go the route of dedicated hardware for HAOS and at some later moment another unit with Proxmox and all the other appliances, like torrents, media streaming, and network helpers, (maybe even a HAOS backup server) on that other unit. That unit might also be a bit more capable than the one that will run HAOS and it will help reduce electricity consumption because all the VMs and Docker are currently running on my main PC.
I switched to an Intel NUC i3 about two years ago- it has been the best decision for Home Assistant. I am running native HAOS Generic x-86 image) as this server is exclusive to Home Assistant. It’s remarkably stable and no docker management needed. I have a 500GB M.2 main drive, but only using 4% of it, so overkill. I don’t recall how much RAM is in the NUC, probably 32GB, and probably overkill.
Now a restart takes 15 seconds. Longer for core updates. Node Red deploys take 5 seconds. A snapshot takes about 30-seconds.
One I have set my target on is a second-hand one, NUC7i3BNK based on 7100u 4GB with 128GB SSD and power supply it goes for 120EUR. The one mentioned before got sold before I could act, that one had i3-5010u, 8GB, and 120GB for 110eur. New ones based on J4005 with 4GB are starting at 170eur, and in cheapest version have 320 or 500GB mechanical drive.
Up to recently, I did not even know that a NUC is … what it is. But, it is a convenient little device from some perspectives, a laptop without an integrated display and keyboard, a space-saving, and a power-saving device. I would also assume a nightmare when the motherboard dies, similarly to laptops.
From a B2B perspective, you can purchase an integrated third-party UPS which is nice for 100% up-time. When managing a fleet, it’s easier to manage an image with the same architecture (e.g. AARCH64) across all devices as opposed to having to manage multiple images.
I have three NUCs in my basement (on a UPS). One is Home Assistant, one is for Plex and the last is running Ubuntu for those things that Linux does better than Windows.
They have been running for two years without a single moment of problems.
There is also a Raspberry Pi3 which is my MQTT broker and a Viewsonic VOT133 (Point of Sale) PC that is a small NAS. The Sonic is slow, but it runs Ubuntu and only cost $35 on eBay.
I started thinking in a similar direction, one for HA, one for media-oriented containers, and one for security-oriented containers. Taking into account that they are all DC-powered, UPS functionality can be made without the device that needs a set of new VRLA batteries every 2-3 years.
IMO way too many computers. Depending on the specs of the NUCs, you could either run everything as a VM or docker container or both. I personally think, that going for one higher-power PC instead of multiple lower-powered PC´s could save energy.
When I used my old office pc from 2006 as HA server, I had no issues also using it as samba server. Maybe it was a bit slower, but it was perfectly fine back then. My new server, which is my old gaming rig, became then my server where I run everything on. Host OS is ubuntu server 20.04 LTS, HA runs as supervised (i dont care about official support, im a tinkerer anyways), some addons are also used, then some additional docker containers, also samba server, it acts as my backup repo and my backup softwares server relies on windows, so it runs in a win10 vm. No problems at all, I believe I could also add some AI things.
Since I had to lend my servers PSU to a friend and staying in a temporary apartment, I use my old NUC as server lite, means only HA again as supervised under ubuntu server 20.04 LTS, some addons and some other containers, which still runs fine. Ok, it doesnt run a VM and doesnt do file serving, but still its enough to handle my automation needs while still not consuming much power.
If you’re simply running HA, most NUCs are overpowered. And the low end NUCs have a lower passmark score. Unless you’re going to do some video processing or share the box with another service.
It all depends on the needs, and the location of a particular user I suppose.
I finally got the first NUC yesterday. It is based on i3-6100u. It does not have stellar performance, but at 90eur, I would rather have 2-3 of those and have some redundancy if one fails than have one machine running everything, costing 4-5x as much as this one, and losing everything if the machine should die unexpectedly.
My primary pc gets sold off for about 60-70% of its initial value after 4-5 years of use (not the whole device, just the set mb+proc+memory). And I always value that when the time comes for some components to die, it happens in someone else’s ownership. After having a 3yr warrantied power supply die after less than 4yr twice I have learned that it is more practical to buy a 5, 7, or 10 yr warrantied supply, sell it off with a warranty going another year, and buy a new one than waiting for it to die in my care.
Now when I have a working NUC to play around I will install Proxmox, a few servers, and HA also and see how everything works. Learn the nuances of Proxmox and containers, and at some point, if I like the experience add another NUC for other containers. Then I will have the opportunity to power off my main pc when it is not used. This will in the end bring the power consumption down.
glad to find this thread, had just opened this myself:
running a mini PC with Celeron for 2 days now (upgraded form RPI4 which started to struggle), in doubt whether I should upgrade a bit further, maybe I3 indeed… but also considering power, which I dont see mentioned here?
