Large Scale Residential Running Home Assistant | Home Assistant | Page 1

ashen yoke Nov 14, 2024, 11:32 PM

#

I'm building a 7,000 square foot home with plans to have whole home audio (12 zones), whole home video (6 zones with TV, AVR, streaming devices like Roku, etc.), smart light switches throughout, hard-wired sensors (e.g. motion, door, window, water leak all tied to an Elk M1 Alarm Panel), at least 1 voice assistant per room/area (exact device still TBD but leaning towards FutureProofHome's SAT 1), 4 garage door sensors (most likely ratgdo), and lots of smart shades/HVAC control/temp/humidity/etc sensors. BTW - I built a similar home 10 years ago using Universal Device's IS994 home automation platform and it was nearly flawless.

Could anyone that has a large smart home (e.g. or home with hundreds of "entities") please share your experience running Home Assistant at scale and include what hardware HA is running on, how many entities do you have, and/or anything relevant?

mental obsidian Nov 14, 2024, 11:58 PM

#

1500 entities here in my HA instance, much smaller house too, I run it as a vm on an x86 mini PC, really the add-ons are what eat up ram on my system, music assistant taking 1.5 gb worth.

jovial grail Nov 15, 2024, 2:02 AM

#

I have ~5000+ entities (400+ "devices"). No issues at all. HA Container is running on a N100 MiniPC (16GB RAM/Alder Lake CPU) with a Unifi network (UDMP, 3 APs, 6 switches, etc). Granted, small house, but nearly everything is "smart" in some way, shape or form.

mental obsidian Nov 15, 2024, 2:30 AM

#

for me, the main thing i have to be sure of is making sure my wifi is stable and solid everywhere, more lower powered APs rather than one big one

#

also depending on scale, i've heard that zigbee/zwave/matter all can get problematic when you get into hundreds of devices on the networks

jovial grail Nov 15, 2024, 2:33 AM

#

You can counter Zigbee/Zwave issues by having a good coordinator that's connected via LAN and not USB coordinators (like Tube's CC2652P7 or MGM24 or SMLight's models (P7/P10/MGM24)) can handle upwards of 300 routing devices with each routing device handling 5-10 children each (so, ~3000 total devices). You can also do things like multiple Zigbee meshes as well. ZWave is more forgiving in terms of devices and range of course.

mental obsidian Nov 15, 2024, 2:48 AM

#

i assume you would do this anyway, but also lots of ethernet backhaul, since its that big too, might as well have them install the ceiling ethernet drops for the APs, getting as much as you can hard wired instead of wireless will mean the wireless is better for the things that can't be hard wired

ashen yoke Nov 15, 2024, 3:20 AM

#

mental obsidian i assume you would do this anyway, but also lots of ethernet backhaul, since its...

Yep I’m pulling my own low voltage wiring and will have many AP’s for great WiFi coverage

tribal prairie Nov 15, 2024, 8:43 AM

#

Z-Wave caps out at 238 (from memory), ZigBee doesn't, but needs a good coordinator (and other device choices) to still be performant at that point. You'd probably be looking at multiple meshes past that point.

ashen yoke Nov 15, 2024, 5:50 PM

#

@jovial grail @mental obsidian @tribal prairie - Given that I will have many devices (and lots of HA entities), coupled with running various add-ons and integrations (e.g. Elk Alarm Panel, whisper, piper) etc., what hardware would you recommend I use to install HA? I will have an A/V rack in my basement so a traditional server is okay. I also plan to use a local LLM for private voice capabilities so I'm not opposed to purchasing a beefy server running Proxmox (or similar) and allocating resources appropriately (e.g. container for HA, container for LLM, etc.)

jovial grail Nov 15, 2024, 5:54 PM

#

ashen yoke <@534201530895564800> <@214942457375227916> <@332579266623438848> - Given that...

The HA server itself could be something like a Mini PC. HA does some of the "heavy" lifting, but it's still very lightweight in terms of specs. A N100 Mini-pc should be more than plenty to run it.

#

But, for the local LLM, you're going to want something good. I just got a Mac Mini M4 yesterday and put Ollama on it with Phi, Gemma, and Dolphin-mixtral and tbh, it's blowing away my beefy server in terms of token speed and inference. Plus, at $599 for the base, it's a steal.

ashen yoke Nov 15, 2024, 6:06 PM

#

jovial grail But, for the local LLM, you're going to want something good. I just got a Mac Mi...

That's a brilliant idea! I assume you got the Mac Mini M4 with 10 core CPU and 10 core GPU and 16 GB of Unified Memory? Approx. what response time are you getting when using a specific model? If you had the $, would you have purchased a bigger Mac Mini (either with more memory or cores)? thank you!

jovial grail Nov 15, 2024, 6:32 PM

#

Yup. I wanted to test it with just the base model. I'm not sure I really need the beefier Mac Mini. This is from running a fairly complex prompt.