Hey,
I am sorry if this question is already asked. New on the forum and tried to search, but could not find the answer.
I am buying a house which i am going to control with home assist. I plan on having speakers mics in each room. Prescence sensors, a camera in the kitchen to do picture recogniction of fridge content etc.
What hardware should i go for not to have to much voice lag and be able to handle camera and object recognition?
Hardware for the home assistant server.
NUC
Frigate uses coral
As much ram as available
OR
Used PC / laptop
Use for a while then buy when you understand needs better
Is an i3 laptop enough to get started? I was “gifted” an i3 with 8GB of RAM and and 240GB SSD. No voice control nor cameras for me. I have Blue Iris running on its own dedicated machine.
I installed HAOS on a Lenovo Mini PC (ThinkCentre) that I got off of eBay for $40 USD, which works quite well. I was not able to find any minimum system requirements for HAOS x86-64, but this thing will run on a raspberry PI so I assume that you would have to really dig to find an x86 mini pc that won’t run it.
I also bought a Sonoff zigbee dongle (SONOFF Zigbee 3.0 USB Dongle Plus | ZBDongle-E) to integrate my zigbee devices and it works really well.
Thanks @amobley I’ll likely give it a go over the holidays. I’m currently running HAOS in a VM under VirtualBox and I keep getting told that’s a bad idea. Long term I’m looking at Proxmox VE but I think I need some hefty hardware for that.
2GB RAM, 32GB Storage
I also found 64-bit and UEFI boot capable but nothing specific about processor. I’m sure that depends a lot on what integrations and add-ons are running.
I have heard plenty of people running HA in Docker and few in VMs. I am sure there are advantages, like not having extra hardware, but to me, that seems like a lot of unnecessary complexity and fragility when this thing can be deployed on a rock solid mini pc like the ThinkCentre which can be had for $40 (not counting the zigbee dongle).
I don’t consider myself an HA or Containerization/Virtualization expert so take that as you may.
Could you tell me where this is documented so I have it for future reference?
I did. Well, it would run but the latency would be horrendous.
It is a Zotac micro PC (meant for cash registers) running an AMD G-T56N Processor and 8Gb of RAM. The slowest computer I’ve used since my Windows 95 PC. But it’s fine for running RTL433.
OP- You can buy a used Intel NUC i3 or i5 on eBay for less than the price of a new Raspberry Pi5 plus case and PSU, and the NUC will outperform the Raspberry Pi in every metric.
I run my Home Assistant server on a NUC i3 and until recently it also ran the Frigate Add-On with nine IP cameras. But the Add-on was consuming most of my server’s RAM and would occasionally crash it, so I moved Frigate to its own Intel NUC i5. (I probably could have just purchased more RAM).
I am strongly opposed to complicating HomeAssistant with Proxmox or virtualization. If you must use the PC for something that there is no Home Assistant add-on available, you can always complicate your installation with Proxmox later. To start, just run Home Assistant bare-metal. Subscribe to Nabu Casa and do nightly backups.
I haven’t ventured into the voice control. Yet. For now, I am pleased with having an Alexa Echo device in every room. There are several ways to integrate it with Home Assistant; most use Nabu Casa. I use Node Red.
Fridge content? How does a camera, likely in the ceiling, see the mustard behind the mayonnaise? Why not just buy a fridge with a camera built-in? (I saw one in Lowes a few years ago).
The free version of Frigate has rudimentary object recognition. You can train the paid version, but I haven’t ventured into that. Yet.
Free version has local training for object recognition. Allows you to identify open / closed door and likely other things
License plate recognition as well
No GPU needed but Google coral required
How? I wanted to know when an Amazon box was on the porch but couldn’t figure out how to train the free version.
Interesting- any documentation on this?
I’ve been using HP Thin Clients, the HP T730, then T740. Can get them from eBay for $100-200, and the T740 at least comes with a 2000 series Ryzen processor and integrated GPU. Has room for a Coral, 2 NVME SSDs and 10GBe network card too.
Feature was added in recent release
https://docs.frigate.video/configuration/custom_classification/state_classification/
Not at all.
Proxmox is pretty light.
The question is if there is hardware resources to spare for extra VMs or containers.
An Intel I3 and 8Gb ram could run 2 VMs in Proxmox without too much issue.
+1 on Proxmox not being massively hardware hungry when you’re on the sff-pc scale of things. We’re running the haos install as a VM on an old mac mini or a small 2nd hand lenovo box dependent on where high availability landed it.
i went off VM recommendations.
There was an actual recommendation in past but it got removed at some point.
Thanks @WallyR and @AngryAnt The current SSD has Win10Pro on it so I’m going to swap in a blank SSD and give Proxmox a shot.
@stevemann I’m looking at Proxmox to ease the backup process using both backups and snapshots. Spinning up a clone for dev testing is something else I’m looking for. There may be other ways. I’m not set on anything at this point.
In the past couple of weeks I just purchased 2 HP ProDesk 600 G3 with 8Gb of memory and installed HA bare metal in about 10 minutes with a live Linux USB and dd command. Quite nice machine, very sturdy, easy to work on. I’m giving them to a couple of my grown kids for Christmas with the new ZBT-2 and a pack of Zigbee smart plugs. I doubt they will ever have a keyboard or display connected.
Very nice. I purchased a ZWA-2 and so far it has worked well. I plan on getting a ZBT-2 at some point but the Conbee II is working fine for now.