Best/easiest to support hardware/OS/architecture for a powerful hass+tensorflow deployment

Hi all,

I’ve recently moved to from OS X to Hassbian to run Tensorflow. It’s pretty cool functionality, but when updating only every 10 seconds, it’s not very useful for any actual use in security situations. I feel a refresh rate of about 1 per second and one of the more powerful training models would be better. For this I’d need new hardware.

What do you all think the easiest, most supported, most reliable, and most powerful architecture is for hass? I want to have RAID disks and an intel CPU, maybe even a cheap Nvidia graphics card to run tensorflow on CUDA.

I have considered taking one of my old mac minis and installing Ubuntu and running everything as dockers, but threads like this don’t give me great confidence.

I’d like to have the benefits of hass.io as it seems to be getting a lot of the juice these days, and it appears that there is some chance to run it on a generic linux server, but that seems to be limited to very select hardware, none of which is very powerful… It also seems a bit silly to buy another new machine for this when I have a bunch already lying around.

Or is there some way to run Hass on the Pi and offload the Tensorflow to another box? I am getting a powerful new mac mini and it would make quick work of this.

Any ideas? I’d love to know what people with more powerful (non Pi) Hass installations are doing.

@scstraus - have you made any progress on this? I’d love to hear what you’ve done. I’d like to use HASSIO & Tensorflow together on my intel based VM.

Yes. I’ve gone to Hass.io installed on a Ubuntu Server on an old Mac Mini. For tensorflow I am running the Frigate Tensorflow to MQTT bridge in a docker container on a VM on another machine (currently using CPU for detection, but will soon move to Google Coral). Great architecture, I highly recommend it!

More about Frigate in this thread

1 Like