The Coral TPU driver has basically been abandoned by Google so if you are running a Linux kernel newer than 6.2 it will not function.
https://github.com/google/gasket-driver is the original driver which was archived on April 18, 2026
You can try the driver https://github.com/feranick/gasket-driver or https://github.com/dude84/gasket-driver-coral or search through the forks of the original gasket-dkms driver https://github.com/google/gasket-driver/forks
So in the future your options are to pin your kernel to 6.2, upgrade your hardware, hope that someone will keep a gasket-dkms fork updated for newer kernel versions, or make your own fork to do so yourself.
You must log in or # to comment.
Technology websites should just add a top level menu - “Google Abandoned”
Google is running out of naming schemes for their projects.
Not enough words in the dictionary.
even though they have the full alphabet??
Man, why did they buy Tenor just to kill it?
What the fuck.
Common pattern - the acqui-hire.
“These people are working in a problem area that we want to do better in. We’ll buy their company for their expertise.”
Whether they keep existing products or not is not a major factor in the decision and gets evaluated later. Often, because they want the people working on something new the existing products are put into maintenance mode or shut down.
Source: Have been acquired for both talent and for product. Seen both.
So they could sell you something else, duh…
P.S. They probably wanted to add the IP to their portfolio.
As much as you may dislike Google, I got to hand it to them, they have and always have a ton of skunk works projects.
It’s only Google
Great, I bought one like 6mo ago and have it running on frigate…
<sigh>
The frigate container comes with the drivers for it. So its probably fine for the time being
I wonder, is it due to architecture limitations? They don’t see a roadmap ahead for it anymore in light of changing AI hardware demands?
I assume the hardware is end of life and not just the Linux driver.
I’m guessing it’s because more powerful hardware is coming out all the time. But for a lot of homelabs more power isn’t really needed to watch a few cameras for basic detection.
And yes the hardware hasn’t been made in a while but new old stock is still being sold. Hence the reason for the post.
For a few cameras with basic detection, an Intel 6th gen processor or newer is sufficient to run openvino on the iGPU. Works great. https://docs.frigate.video/configuration/object_detectors/#openvino-detector
This is great, if it works, on my 8500t I had nothing but problems.
I use it on a 12th gen CPU and it worked first try. ¯\_(ツ)_/¯
Yeah it seemed to work for me and would then hard lock the system between 24h and a week later requiring a manual power cycle. Several people experienced the same, we had a couple of issues on the frigate github but I don’t think anyone ever figured out why. Definitely openvino on the igpu though, openvino on CPU was fine.
What is a good m. 2 alternative? I was looking to use my old rx580 but it appears that rocm dropped support for it
Perhaps the RX580 is usable via Vulkan? I tried Vulkan with llama.cpp on a R9700 recently and it was generally faster than ROCm.
Well, good to know. I planned to buy one and attach it to my homeserver ಠ╭╮ಠ
I think this plan needs to be replaced.
Nowadays just get an Intel Arc A380 for 150€ and you can use it for a lot more than only Frigate. That thing is a little beast for my server.
Unfortunatly this is not possible for my setup. I have a Asrock J4050 with its soldered cpu. So I need something like a separate device. USB, SBC with lan, anything like that… Because of that, I was happy when i discovered coral. And to be honest: I did not researched deeply. So I am unaware of its limitation to frigate. I wanted to create my own model and run it on coral. The training was planned to be done on another device.
Now I need to check if a raspi could be sufficent or if anything else comes up.
