Should I use nuc instead of pi?


#1

Hi! Home assistant is getting more that some fun thing for me. I am going to leave second system alone at my parents house, So I need it to be solid stable. Here is what I have:

  • 7 z-wave devices
  • 22 xiaomi/aqara devices
  • Around 700 strings of automation
  • Couple cameras (only mpeg on HA)
  • Weather, launch info, some small things
  • Two small add ons (samba, nut)

Home Assistant seems to be just fine, as processor usage is on 1-3% (30% on reboot). Memory is on 20% (16 gb card). But I am a bit worried because of others topics on cd card issues (it does not like often writes in database and suffers from power lose).

Am I paranoid? I even have UPS connected now and I am getting notifications when electricity is gone. Will it be enough in my case to make an automation which would peacefully turn off pi when UPS is going to die? What if I will have more aqara sensors and switches (40). Will pi still feel fine?

And actually, I did not notice any issues for now, but I am always doing something on pi and max time I did not reload it was 1 month.


#2

When you make an automation to turn off the Pi, please share it here.
I am surprised what the Pi can handle- keep in mind that most of the ‘stuff’ you have on HA is not running all at the same time. I have three cameras on my HA, and those do bring the Pi to a crawl when they are on-screen.
On the SD card- don’t be cheap here. Go for a 64Gb, Class-10 micro SD, even though the Pi will run fine on 8Gb. Here’s why. The more expensive cards have a feature called ‘Wear Leveling’. Here’s a White Paper from SanDisk: https://web.archive.org/web/20150326122100/http://ugweb.cs.ualberta.ca/~c274/resources/hardware/SDcards/WPaperWearLevelv1.0.pdf
If a card has wear-leveling, then all writes are going to be to lesser-used sectors first. With a 64Gb card, you have a lot of unused space for the wear-leveling to write to before you ever use a sector for a second time.


#3

Hm very important info on 64 gb. Sure will get one (now it is 10a class but, as I wrote - 16 gb)
When I had rtsp live camera - pi was not happy at all (50% pocessore use). So now I use DVR with another app for that
And yeah, thank you a lot.


#4

From my own experience I agree with what @stevemann has said in regards to SD card. I found however when I started to have a few sensors that the cpu cycles were busy performing those tasks and automations would suffer from responsiveness. I found moving my automations to a standalone Pi improved automation performance immensely. I have however wanted to do facial recognition and the Pi is not up to that so when my latest SD card failed I decided to move over to an Intel Nuc and wow man firstly restart times are less then 2 mins but the whole thing is just snappier the UI is responsive I have moved back all of my tasks being run on different Pis to the one ubuntu server and now with magic mirror and facial recognition. I recommend it if you can justify the extra cash outlay for it.


#5

Cool! And did you use tensor flow for face recognition? I would love to have face id in future. For now have more simple tasks


#6

I want to continue using my Pi because I have a number of wired Reed Switches that I don’t want to replace with something else.

To be safer, I switched a little while ago to a high endurance mSD card:

And just over the weekend I replaced it with an SSD that my RPi3 now runs on now - little bit of work, loads of learning opportunities and, hopefully, a more reliable setup:


#7

I’ve gone from pi to Intel nuc and I recommend it


#8

I am running on a pi at the moment but I have an option of buying a gen6 i5 with 32 gb ram and ssd. I dont know if gen6 is fine but I hope so and I will probably go down that route. Currently running hassio on pi. Want to use the nuc for more than HA so guess I will run home assistant in docker or something. Want to be able to use tensorflow!


#9

Gen 6 i5 would be more then enough, I am currently running an i5 gen8 and its massive overkill cpu never climbs above 5%. I installed Ubuntu server (headless) and run HASSIO in docker so I can continue to use the addons.


#10

Nice, thanks. I want to use it for radarr, sonarr etc as well. Hassio is great but yet doesnt support tensorflow. Hopefully there will be support in the future.


#11

I use machine box on the same server here is testing with facial recognition yesterday cpu spike to 10%

image


#12

Thanks, how many watts does the nuc draw?


#13

I think any Intel nuc would be great. If you are not running Plex as well then the cheaper the nuc better probably. I think I am using an i3 or i5 5xxxx and it is easy over powered. Over the last 24hr it’s been on 1-2% except for a transient spike to 8%. It’s running hass.io with about 10 add-ons (unifi zigbee2mqtt mosquito samba SSH and some others).


#14

If you do ever want to move to a NUC you leave the wiring to the pi alone and just install this https://github.com/flyte/pi-mqtt-gpio on it to send the sensors to your new HA server via MQTT.


#15

Good point - I actually have a NUC, but it’s running my NextCloud installation at the moment :grin:


#16

I don’t currently monitor the power consumption of my server rack as it might scare me to know what it costs me to have my routers, switches, NAS, Nuc and PC running 24-7. I am waiting for a efergy energy consumption monitor to arrive and I will start to concentrate on what is my biggest power consumption culprits.


#17

Ha ha! Roger that!


#19

My NUC (NUC7PJYH2) running homeassistant, Elasticsearch, Motion (4 cams) and a bunch of other programs consumes on average 6 watts. My entire IT setup (router, 2 raspberry’s, modem, Hue bridge, Synology NAS, power meters, …) is about 55 watts which is a bit too high to me.


#20

Do you ise nuc like a DVR?


#21

What processor for your nuc? Was planning on adding facial recognition myself. Also, is your nuc running cameras or is that on a separate system?