Bare metal + frigate add on + coral config

Hardware
1

Looking to see if someone can advise me on this before i jump in. I have decided on changing how i run HA (see here Explain to me NUC, bare metal etc please! - Hardware - Home Assistant Community) and have settled on bare metal, running frigate and plex as add ons.

If i add a coral for frigate, with HA being bare metal, does it just ‘automatically’ know its there? I’m considering either USB or PCIE for the coral, if I was on windows i’d expect to be installing a driver for the PCIE version so does HA/Frigate handle this?

2

I’m using a Coral (usb) on “bare metal” HA on a NUC. When you install the Frigate add-on, you need to add a line in the frigate.yaml file mentioning the Coral (see Frigate documentation).
There’s no further need for installing drivers or the like.

3

The PCIE coral did require drivers to be installed separately in the early days. That may or may not have changed.

4

Google Coral TPU doesn’t require a separate driver, only configuration setting in the frigate.yaml file.

5

OK - are you using usb or the pcie version? I thought the usb might be plug and play, wasnt sure about the pcie. Will check the frigate docs again, i dont remember seeing it when i read through…

6

Almost. Plug it in, tell the config what port it’s on and restart Frigate. Done.

1 Like
7

Same as @stevemann . Works like a dream

1 Like
8

How many cameras are you guys running? I have 12 on the system so there is a few to go at, i feel like the USB version will still be fine - the pcie is just a little cheaper or more powerful depending on which one you go for.

9

To be clear the PCIe version of the coral does require a driver to be installed on the host. HA OS includes this driver by default

10

I have eight cameras running RTSP protocol.

1 Like
11

PCIE inference is about 10 per tpu
I think usb coral is about 18

From here

The USB version is compatible with the widest variety of hardware and does not require a driver on the host machine. However, it does lack the automatic throttling features of the other versions.

The PCIe and M.2 versions require installation of a driver on the host. Follow the instructions for your version from https://coral.ai

A single Coral can handle many cameras using the default model and will be sufficient for the majority of users. You can calculate the maximum performance of your Coral based on the inference speed reported by Frigate. With an inference speed of 10, your Coral will top out at 1000/10=100, or 100 frames per second. If your detection fps is regularly getting close to that, you should first consider tuning motion masks. If those are already properly configured, a second Coral may be needed.

12

Excellent - exactly what i needed to know, thanks!

13

One last question then while i have the ear of people who might know!

If i go pcie, i’m not sure which card to get. Info on the motherboard is a bit scarce as it looks to be a proprietary HP job, info at the link below, pdf pg 22, comes up as HP83 E1 in the system info on W11. I cant see anyway how i would know if the bus is dual channel so the dual coral is probably out, so then it comes down to the interface. Looks to me like i probably need the standard ‘mini pcie accellerator’ into slot 2 or 3 on the board?

I’m not sure if its relevant as they are not listed as ‘expansion’ but I have installed a NVME HD on slot 14, 10 & 15 are free.

Maintenance and Service Guide HP EliteDesk 800 G4 SFF Business PC

14

Dual TPU in single channel slot will result in only one tpu working. I did this initially and it ran fine. Honestly a single pcie TPU can process about 100fps(infererence speed 10) and at 10fps/camera that’s 10camera. If you reduce to 5fps/camera you’ll be OK for up to 15 cameras I would presume. I am no expert just guessing.

I eventually got single pcie to dual adaptor which allowed both tpu to function from my unit. Now both can process about 100fps each or 200fps total

Pretty sure this is the adapter I got along with the m.2 tpu. This is the projects GitHub page. There is Q and A on GitHub page that explains the dual / single problem well and has info that you will find useful.

1 Like