I run HA on an RPI 4B (2GB). I have no issues with it, and I am in no way pushing any limits in terms of storage or CPU use.
I plan on adding an NVR with 4 cameras to my network with feeds to HA, and also it seems that functions for speech to text will be hardware demanding. Also, any AI for image recognition would be demanding. With this in mind I have been looking at two small form factor (SFF) alternatives:
1: I5-8400, 32GB RAM, 128GB SSD
2: I7-6700, 32GB RAM, 128GB SSD
Both can be expanded with a GPU in the future.
My questions:
Which of the two options would you pick, and why?
Would it be sufficient?
What issues can I expect? I am backing up HA to google drive, and my thought was to just download from google drive to the SFF.
I started out running home assistant on an Odroid HC2 but it kept blowing it’s boot flash drive so I upgraded to a off lease HP SFF I7. The I7 running home assistant as a docker container (with 9 other containers) under linux used about 0.02 of the CPUs as reported by htop and comsumed 30 watts of power.
Since the Ukrainian invasion jacked electricity prices I’ve become very fond of thin clients.
I currently using a Wyse 5070 with an Intel Celeron J4105. 4.92 watts 0.24 cpu utilization running 12 containers.
Warning on Frigate without a TPU: it was drawing 28.1 watts alone! or $104 per year at my utility rates!
edit: that was frigate running on the I7
#2 if cost is not a factor #1 if huge cost difference
Most services require very little cpu or memory. Plex and NVR being the exceptions. In case of NVR ffmpeg is commonly used for
Decoding/encoding video or other need and this can be cpu intense. I use frigate which uses little cpu mostly because it is offloaded to TPU or GPU. Recording is not much of an issue, its the motion detection or object detection done by any NVR. If you dont do that, have GPU or TPU for objects it is not an issue.
Plex has similar issue. Any transcoding needs to use cpu but an i5 can handle this well as long as you dont have several 4k streams simultaneously running. A single stream CPU alone is no problem. 2 depends on several things and generally not a concern as this can also be done by streaming device in some setups.
1 and 2 will improve HA experience. Especially,again, with camera viewing within HA
The cost difference is minimal. Are there performance gains by going for the I7? I googled “I5-8400 VS I7-6700”, and from the results it seems the I5 is better. I assume because it is newer. Something I am not seeing perhaps?
Update: The i7-6700 is a " i7-6700vpro" according to the listing, but the I5 listing does not mention “vpro”. One is a “ThinkCentre” and one is “Elitedesk”. Can I assume both are “vpro”, and does it matter?
I saw some minor benefits that favored it. I dont trust benchmark comparisons between chips so similar. Some say i7 better others say the i5 depending on what you using it for and for HA both will be great. Even power use i expect to be based on use. Ive found passmark to be good rating of how well video decode/encode will perform but again, too close to matter.
I think more threads (i7) mean more simultaneous process and faster clock (i7) mean faster processing and again, specs are to close to care. I would just get cheapest or best availability.
I found an even better option and bought that one:
Ryzen 5 3600
16GB DDR4 RAM
512GB M2 SSD
A Nvidia GPU, model unknown. Based the serial number on the box label it is likely a GTX1650.
The seller did not appear to have in depth knowledge about the computer, but from what I saw I was happy enough. Ended up paying 189USD including shipping for it, which I think is a bargain.
Seems Proxmox is the way to go, as PecosKidd suggested. I have no clue what to expect, but HA tutorials being as they are I think it will be fairly easy.
Anything special I should consider/ keep in mind before installing Proxmox, considering my intended use with the NVR and Frigate?
Will a coral TPU be required? Or just more efficient? About power cost, I will have it placed where a few extra watts of dissipated power is welcomed.
Edit: Google answered this for me.