NUC gets my vote
I did 3 year ago Had no problem with it
I would say at least a small PC possibly running unraid with HA & Plex in docker.
It would be cheaper this way than getting an external GPU. There would be plenty of headroom for person detection, space as storage requirement grows.
This will also allow headroom for any tinkering/testing of other projects/containers.
I used to run everything on unraid and have HA in a vm for portability but ended up with unraid running Plex and everything else I need and now HA has it’s own nuc so that I can take the main server offline for maintenance and not have to spin down HA.
Ideally Odroid or Nuc for HA then it can be completely independant of other services/downtime and run Plex on something that is cost effective to upgrade without needing external hardware like eGPU, it will be much more reliable too.
Agreed, with one addition: Purchase a used Celeron, Pentium or Core i3 NUC from eBay. Install a good quality m.2 SSD stick and at least 4GB RAM if not 8GB RAM. You’ll never want for reliability or performance again.
1 Like
I started with RPI 3 and SD card, had no hardware issues in a year of running. However all the conversations about SD card failures made me decide to get off that. It was frustrating that there is yet a good cost effective rpi 4 with m2 storage option. So I went with a support expired Mac mini with Linux running and docker. Very happy, moving to intel hardware is still a more complete universe than arm/rpi. The apple hardware is not perfect, there are some quirks, ethernet not super well supported in linux, cpu and motherboard sensors/fans not well supported (although, it appears this is a problem in a number of ‘windows’ intel devices as well). I wanted to have a backup hardware device in case of failure and for testing new configurations. I considered a second apple device, however about this time I discovered the often massive supply of inexpensive windows small form factor machines on ebay. It appears the combination of windows upgrades and on going malware and other virus hacks of corporate and government network makes these entities do ‘rip and replace’ rebuilds down to the mouse and ethernet cable on average of every one to two years, yielding 100,000’s of units in the after market. All with good hardware and software support. Two or three generation ‘old’ intel devices in general seem to be good sweet spot for home automation servers.
So long way to my recommendation, look at these low cost ‘mini’ pc’s on ebay as a good route to explore. I looked at the Intel NUC’s, however these seem be on the high end of the price spectrum, even for used units. These NUC’s seems to be a ‘showcase’ device for the latest intel processors, graphics and i/o technology, and often you don’t need this for a HA headless device.
I am currently using:
HP ProDesk 600 G1 Desktop Mini, 8 GB, 500 GB SSD, Google Coral AI coprocessor
https://support.hp.com/us-en/document/c04240180
Yottamaster 5-bay USB 3.0 HD enclosure
https://mjtsai.com/blog/2021/02/05/yottamaster-4-bay-hard-drive-enclosure/
Have a read/view of this guys on going review of small form factor intel devices:
He is currently recommending, as of this date, because the newer models continue to roll into ebay, the ‘HP ProDesk 600 G3 DM PC Core i5-8500T 2.70GHz 8GB RAM 256GB SSD Windows 10 Pro’, but it is a bit pricey compared to the USD 120 price point for the HP G1’s.
I ended up with two HP G1’s and an extra motherboard for USD 280. Yes I did have some other parts laying around that I used in these, however even the base models I got were 1st class HA servers.
FYI, do read carefully the ebay specs on these used units. They can ‘subtly’ mention that there is no power brick included. You want to avoid these. And there are LOT of different configurations of these mini intel boxes, almost any of them are fine and the pricing is pretty much identical, RAM, storage and CPU are the items that move the price, video, wifi don’t. However, be aware if you compare two different listings, or buy from two different vendors.
Good hunting!
I got this Intel NUC device second hand on ebay for a relatively low price (under CAD200) and it has been running HA flawlessly since. I wish I had made the move from a Pi device earlier.
Specs are Core i5, 8GB Mem, 240GB SSD, feels significantly overpowered for HA alone, but the snappiness is hard to resist. I do not currently have other services running on it other than add-ons within HA.
I was running HA on a pi3b+ with sdcard just as starting, switched over to an HDD via a selfmade usb adaptor and had problems like server crashing and so on. Then i switched to an old office pc from 2006 with core 2 duo, 3 gb ram and threw in a few sata drives (one 220GB for system, one 3TB HDD as NAS Storage), installed manjaro linux (my prefered distro tbh) and used HA supervised install with a bunch addons, plex natively and samba server without any problem. Ok, it wasnt the fastest system out there, but it did the job (i mean, i got it for free from my dad). Then some idiot broke my gaming pc (letting the cpu fall into the socket aint great), but i managed to repair the socket to the point, where i have a stable system running with the defect, that i can only use 1 ram channel, whcih is enough for my server use but not for my gaming and experimenting pc. So i switched from MSI b360 gaming plus mainboard and intel core i5-8400 cpu to asrock x570 gaming phantom 4 and amd ryzen 5 3600 on my gaming bench, got a new case, had intel stock cooler and total overkill 700W psu (even with RGB fan) and an old corsair fan laying around build that into the new server together with my 2 8Gb sticks my ryzen wasnt happy to work with. Still havent migrated all my old data from the nas drive of the old server or plex, but the speed difference is gigantic. Even booting went from 2 minutes to almost instant (ok, the new server uses a sata ssd) and i can finally use the vscode addon and theres no need to keep a vga cable around (sometimes i still have to use the gui or the pc locally)
Can thoroughly recommend the MSI Cubi-N-8GL if you’re looking for a “cheaper” option. I have the Celeron 4000 version, which takes a single 8Gb RAM stick, and is using less than 15% RAM and 5% CPU with my current config. All in (with a 500Gb SSD), was under $200.
I went down the Home Assistant OS route using the NUC image (as it’s a dedicated machine). Obviously if you want it for other uses as well that may influence your decision.
I was also thinking about upgrading my Pi from SD to SSD support, but then again someone said an SSD in fact acts similar to an SD and will most likely die much quicker than an HDD, with all the read and writes of home assistant and its database usage.
If that is true, does it make sense to run home assistant on a tiny PC, such as a NUC, using an SSD, or would it be much more reliable, if a HDD is installed?
I’ve been running Home Assistant on an i3 NUC (purchased used from eBay) with a 128GB SSD leftover from a laptop upgrade for well over a year now. Never any storage issue, whatsoever.
Sd cards are made primary for photos in cameras. So just a few writes, maybe a full delete ones in a while and so on. Running an OS with constant writes and deletes isnt the main thing sd cards were made for. Behind the surface an SD basically creates transistors for a binary 1 and destroys them for a binary 0. Creating a transistor is done at like 3V, but the destruction at like 10V. That wears the substrate out, thus a new transistor cant be created.
SSD´s are made for running OSes on and can handle these constant reads and writes.
A solution for SD wearing out is using higher capacity ones (basically, if it detects a defect cell, it skips using it) making it possible to use the sd longer before it goes corrupt. But SSD´s are much faster.
Thanks for the reply. I was referring to this post, which generally proposed not to attach an SSD to the Pi, but rather move to a NUC, with an HDD.
If the NUC runs on a SSD, I do not see much difference to the Pi with SSD, other than the NUC being more versitile if not running Home Assistant OS and has more computational power, that comes at the cost of higher power consumption.
I had running home assistant on a pi3b+ with sd and on a office pc from 2006 with hdd. The speed increase on the office pc was that huge, that i dumped the pi solution. Also i had a more versatile server with samba, plex, home assistant supervised and addons and the only problem i had was the speed of the vscode addon. Also the ability to install as many drives as i have sata ports is a nice addition.
A nuc with hdd will run faster than any pi, atleast in my opinion. The pi cant use sata, youre bound to usb speeds. Depending on your pi, not even usb3 speeds.
To the power consumption: A hdd uses more power than a ssd.
But ask yourself: Do you just want home assistant? How big is your setup? If youre just using a few devices, a pi is enough. Do you have a big setup with large databases? Then go nuc with hdd. Do yoo want to use a server as NAS, plex, home assistant, database server and so on? Go Mini PC with both SSD and HDD.
My personal setup is home assistant as hobby for just my room, since im living in a shared appartment and my budget is very limited. A few sensors, automations for my everyday comfort and so on. My actual server is my old gaming pc hardware except the cpu cooler. OFC its way overkill, but i dont wanna buy new hardware, if i have semi-defect hardware laying around that i can use.
Hi,
not sure how serious you take this. For example the power consumption. I take this serious, means I have an accurate device to measure this.
Before voice commands the Intel CPU ran at ~3,7Watts. Now with rhasspy on it its like 4,4Watts. (detailed measurement is possible when its really done. As I just played with it.)
But in fact it has like a damn lot more power as the PI. Either way I will also move the log files to an internal HDD instead of the SSD, since I want to have logs of everything (e.g temperature sensors) for more than a year.
So after all my two cents.
- For playing around a Pi is enough. But before its gettin serious, mostly likely you have like 20+ devices in it - THINK about a move.
cheers
fastboot
P.S since i spent a lot of time for research about voice recognition. My Jabra 410 will arrive tomorrow. Hopefully it beats by Blue Snowball. Because the Snowball is crap for this. Especially since when you want to have a clear recognition in a ~big room or even from another room. So lets see how the 410 deals with this. But as far as I could read it can handle it. I am quite curious. 
Seems that “here” it’s difficult to find reliable and cheap T versions of processors, for example I can find an i5-8500 at lower prices than i5-8500T. How much this affect the monthly energy consumption?
Between, for example, an i5-8500 and an i5-8500T under baseline (low) load, there is much difference in power consumption?
I.e., for my lab mini server with HA and other things, should I stay with a T version to avoid excessive energy consumption and “high” bills?
The 8500 has a higher TDP and boost higher. Under low load, both CPUs clock itself down. I personally would take the non-T version, since cheaper and if needed more power. My small home server is powered by an i5-8400 and runs HA as supervised under ubuntu server, some more addons and additional software and a small VM and does that fine.
1 Like
I’m using an HP Prodesk 600 G6 with an i3 10300T,16gb RAM, with Conbee II USB plugged in. It’s also my Plex server (just serving, as all the media is stored on my NAS) and there’s been no problems with it. I picked up the G6 for £270 with a warranty until 2024.
You are lucky in UK…
just have a look on ebay for office pc´s. For small servers those are great, HA itself isnt a huge load, unless you use a VM to run it.
My personal thoughts about home servers: VM´s produce too much overhead, you have to install a full OS on them and so on. Containering like docker gives you seperation without emulating a whole PC. But ofc docker itself aint as secure as a properly configured VM. I run everything I can on my home server as docker container or native, unless I really have to use a VM for windows only software. With this in mind, an i5 of 3rd gen would be “good enough”. Ofc a bit more power than you currently need as recommend, just in case you want to experiment with lets say AI person detection.
Also since I like to tinker around, I use ubuntu server with HA supervised, even if its not officially supported. I mean, debian, the ubuntu base, is supported and theres practically no reason to not try it. When it works with some work done, congratz, you learned something. If not, still congratz, you probably learned something.
My personal thoughts: While there is overhead indeed, it is almost neglectable, and I am not sure about calling it “too much”… especially when it comes to HAOS in VM.
Intel has CPU lines from atom celeron pentium i3 i5 i7 i9. And I run my HAOS in VM on a celeron box, along with various nas related services and OPNsense and Plex and a dozen of other stuff - some in container some in VM… all in one Celeron box.
And my Celeron box comfortably handles those just fine.
And here we are talking about an i5 … which is 3 product lines above what I have. So at this point, I’d say let’s not let the overhead get into the way of your VM or container considerations - focus on other aspects / other factors vs your need / you skillset.