i added a Google AI chip in my Home Server…

Summary

This video script details the process of setting up a self-hosted motion-based recording (MBR) solution using FRIGATE on an unRAID server. The host installs Google Coral TPUs with a Dual TPU adapter to enhance performance for multiple camera feeds, overcoming limitations of a standard CPU. The tutorial covers hardware setup, driver installation, Docker configuration, and troubleshooting, ultimately achieving the utilization of both TPUs for improved surveillance capabilities.

Takeaways

• 😀 The video is about setting up an NVR with FRIgate on an unRAID server.
• 🔧 The user needed a Dual TPU adapter to utilize two PC lanes, which their existing hardware only supported one lane for.
• 🤖 The Google Coral TPU is recommended for FRIgate as it outperforms a high-end CPU for the user's number of cameras.
• 📦 The user initially used an Intel NUC but faced limitations due to the USB port already in use for unRAID.
• 🛠️ The user installed a PCIe adapter to overcome the hardware limitation and successfully integrated it into their unRAID server.
• 💻 The unRAID software center provides necessary drivers for the Coral TPU, simplifying the installation process.
• 🔄 The user demonstrated the process of physically installing the Coral TPU into the unRAID server and configuring it within the unRAID environment.
• 📝 The user had to adjust the FRIgate configuration file to utilize both Coral TPUs for improved performance.
• 🔗 The video mentions the need for MQTT for linking FRIgate with Home Assistant, which the user planned to configure later.
• 🔍 The user encountered some issues with the RTSP path configuration, which they resolved during the setup.
• 🎉 The final result was successful utilization of both Coral TPUs with FRIgate, as confirmed by the system metrics and logs.

Q & A

What is the primary purpose of the server discussed in the script?

The server is primarily used for various tasks including Plex, image processing, and is being repurposed to function as an NVR with Frigate.

What does the term 'TPU' refer to in the context of the script?

TPU stands for Tensor Processing Unit, which is a type of hardware accelerator specifically designed to speed up machine learning tasks.

Why was a Dual TPU adapter necessary for the setup described in the script?

A Dual TPU adapter was needed because the Coral TPU unit requires two PCIe lanes, but the available hardware only supported one lane, hence the adapter to bridge this gap.

What is Frigate and how does it relate to the Coral TPU?

Frigate is a self-hosted motion-based camera recording solution. It is recommended to use with Coral TPUs for efficient encoding and processing, as it can strongly outperform a high-end CPU for the task.

What is the role of the 'Intel NUC' in the initial setup described?

The Intel NUC was initially used to test the Coral TPU unit. However, it was found that only one TPU could be accessed due to PCIe lane limitations, leading to the need for a different setup.

What is the 'Unraid' software mentioned in the script and its significance?

Unraid is a software-defined storage operating system that provides advanced features for managing data, including the ability to install plugins and Docker containers. It was used to manage the installation and operation of Frigate and the Coral TPU drivers.

What is the significance of the 'pcie adapter' in the final setup?

The PCIe adapter is crucial for the final setup as it allows the Coral TPU unit to be installed in the Unraid server, providing the necessary two PCIe lanes for both TPUs to function.

How does the script describe the process of installing the Coral TPU drivers on Unraid?

The script describes the process as involving accessing the Unraid software center, searching for 'coral', and installing the 'Coral Accelerator Module Drivers' plugin.

What is the importance of the 'config.yml' file in setting up Frigate?

The 'config.yml' file is essential for configuring Frigate. It contains settings such as device mapping, MQTT settings, and paths which need to be adjusted according to the specific hardware setup.

What issues were encountered during the setup process, as mentioned in the script?

Issues encountered included initial access to only one TPU on the Intel NUC, needing to adjust IP addresses for the RTSP path, and dealing with errors in the logs related to the real link and VR configuration.

How does the script conclude regarding the setup's success?

The script concludes that the setup was successful in utilizing both Coral TPUs with Frigate, despite some remaining log errors that need further configuration adjustments.

Outlines

00:00

🛠️ Setting Up an NVR with Coral TPUs

The speaker begins by introducing their server, which currently serves multiple purposes including running Plex and other services. They express their intention to repurpose it as an NVR using FRIGATE, a self-hosted video recording solution. The speaker humorously admits to a lack of due diligence in purchasing a Dual TPU adapter without realizing their existing hardware only supports a single lane, necessitating the adapter. They explain the need for the Google Coral TPU, which is recommended for its superior performance over high-end CPUs for video encoding, especially with multiple cameras. The Coral TPU is plugged into an Intel NUC, and the speaker navigates the process of installing the TPU on an unRAID server, including the physical installation of the PCIe adapter and the software setup through Docker.

05:02

🔧 Installing Coral Drivers and Configuring FRIGATE

The speaker proceeds to the software installation phase, starting with the Coral accelerator module drivers through the unRAID app store. They provide a brief overview of the ease of installation across different systems and share a link for necessary packages. After installing the drivers, they verify the functionality of both Coral TPUs through the system interface. The speaker then moves on to configuring FRIGATE, discussing the creation and modification of the configuration file to accommodate both Coral TPUs. They also touch on the necessity of MQTT for integration with Home Assistant and make adjustments to the configuration to reflect the new setup. The speaker saves the configuration and uploads it to the FRIGATE app, making sure to map the correct Coral devices and adjust settings for optimal performance.

10:03

📺 Finalizing FRIGATE Setup and Testing

In the final paragraph, the speaker finalizes the setup by starting the FRIGATE container and logging into the interface. They encounter some errors in the logs but are able to confirm that both Coral TPUs are being utilized, along with the GPUs. The speaker acknowledges that there is still work to be done, particularly with the RTSP path configuration, but is satisfied with the progress made in getting the Coral TPUs to work with FRIGATE. The video concludes with the speaker expressing their success in achieving the main goal of the project and signing off with well-wishes for the viewers.

Keywords

💡Unraid

Unraid is a specialized operating system designed for Network Attached Storage (NAS) devices. It is known for its data protection features and ability to run various server applications. In the video, the creator uses Unraid as the base for running multiple services, including Plex and FRIGATE, demonstrating its versatility and the creator's reliance on it for their server setup.

💡Plex

Plex is a popular media server software that organizes and streams video, music, and photos from a central server to various devices. The script mentions using Unraid for running Plex, indicating the server's role in managing and delivering media content within the creator's ecosystem.

💡NVR (Network Video Recorder)

An NVR is a device that records video in a digital format to a network-connected device. The video's theme revolves around setting up an NVR using FRIGATE on the Unraid server, showing the server's capability to handle video surveillance recording tasks.

💡FRIGATE

FRIGATE is a self-hosted NVIDIA Jetson AI application that continuously records a video stream and uses AI to detect events. The video details the process of installing FRIGATE on an Unraid machine, emphasizing its use for advanced video processing with AI capabilities.

💡TPU (Tensor Processing Unit)

A TPU is a specialized AI accelerator developed by Google that is designed to speed up machine learning tasks. The script discusses using a Coral TPU for the FRIGATE setup, highlighting the need for dedicated hardware to enhance AI processing for video surveillance.

💡Dual TPU Adapter

The Dual TPU Adapter is a hardware component that allows for the connection of two TPUs to a system that only supports a single lane. The creator mentions needing this adapter due to a compatibility issue with their existing hardware, illustrating the complexities of setting up advanced AI systems.

💡mVME Key

mVME Key refers to a form factor for embedded modules that can be plugged into a host system. The script mentions the Coral TPU unit requiring an mVME Key slot, showing the specific hardware requirements for integrating the TPU into the Unraid server.

💡Docker

Docker is a platform that allows developers to develop, ship, and run applications in containers. The video describes installing FRIGATE using Docker on the Unraid system, demonstrating the ease of deploying applications in a containerized environment.

💡Intel NUC

Intel NUC is a small form factor PC known for its compact size and powerful performance. The script initially mentions using an Intel NUC for testing the TPU setup, indicating its role as a test platform before integrating it into the main Unraid server.

💡Configuration File

A configuration file contains settings and parameters for applications or systems. The video discusses modifying a configuration file for FRIGATE, emphasizing the need for custom settings to tailor the application to the specific hardware and use case.

💡RTSP (Real Time Streaming Protocol)

RTSP is a network control protocol for streaming media over the internet. The script mentions adjusting RTSP paths, indicating the process of configuring video streaming parameters for the FRIGATE application to function correctly with surveillance cameras.

Highlights

The server is being repurposed for an NVR with FRIG, a self-hosted motion-based recording solution.

A Dual TPU adapter is needed due to a lack of proper research before purchasing hardware.

The Coral TPU unit is recommended for FRIG due to its superior performance compared to high-end CPUs.

The server initially had issues accessing both TPUs due to hardware limitations.

A PCIe adapter is used to overcome the single-lane limitation of the server's hardware.

The installation process involves physically installing the TPU adapter into the server.

Unraid software center provides the necessary drivers for the Coral TPUs.

The server's storage setup includes a parody drive alongside other 4TB drives.

Both Coral TPUs are successfully detected and operational after installation.

FRIG configuration requires creating a folder and config file for the setup.

MQTT is necessary for linking FRIG with Home Assistant.

The FRIG Docker container is set up with specific configuration parameters for optimal performance.

Multiple devices can be used for object detection, including the Coral accelerator and an Intel GPU.

The FRIG setup process involves adjusting IP addresses and RTSP paths for compatibility.

System metrics show GPU usage and interface speeds, indicating the utilization of both TPUs and GPUs.

Despite some errors in the logs, the installation is considered a success with both Coral TPUs operational.

The video concludes with a summary of the process and a call to action for further configuration adjustments.

Transcripts

hey there this is a server this thing

does a lot for me is currently my unraid

machine right now I'm using it for Plex

I'm using it for image I'm using it for

a lot of things but what we're going to

be using it for now is an NVR with frig

and I've been meaning to do this for

quite a while but I needed to wait for

this guy this right here is a TPU

adapter now why do I need an adapter

that is because I am an idiot and I lack

the common sense to do proper search on

things before I buy

them there it is it is the Dual TPU

adapter and I have this little guy this

is a coral little teeny tiny dual TPU

unit specifically this goes in a mvme e

key and the reason I need an adapter is

because this requires two PC lanes while

all of the uh e keys on any of the

hardware I have only support one lane

this right here fixes that I'm getting

ahead of myself let's add some context

here this right here is the frig website

if I go to recommended Hardware it goes

over everything that you're going to

need to get this set up what this is

like I kind of said is a self-hosted MBR

solution allows you to record um there's

different snapshots for different types

of events and just relying on something

like a Intel CPU so for Intel Hardware

encoding does work but for the amount of

cameras I have does not work very well

we can see the interface speeds here if

I just had a couple cameras set up i'

just use a little like n200 Nook but I

have quite a few this right here the

Google Coral TPU is the recommended

thing that they uh recommend you can see

here it will strongly outperform a

$2,000 CPU and it will work with any

Coral device the most common one that

people use is a USB but unfortunately I

already have a USB hanging out of this

thing for unraid and that is uh going to

look a little messy so I got this thing

uh plugged it into a Intel Nook

initially and I only had access to one

of the tpus which is why I needed this

this I cannot plug into a Nook this is a

little pcie adapter but fortunately this

right here this is the uh ug green uh

dxp 6800

Pro this thing over here I think right

here has a spot for me to install this

little pcie adapter and within the ungr

ungr unraid software Center they do have

the drivers so I haven't tested this I

did test this on the Nook works great I

did it with home assistant but we're

going to be installing it on here with

Docker so the first part of this

installation process is obviously

installing it so we're just going to get

this flight it on in and screw this down

there we go I wish that was the easiest

part now I got to take apart this thing

pop out all these drives to make this

easier to move around look at that

casual Flex casual Flex that's the only

one that's my parody Drive everything

else is 4 tabt with the unrated you

can't have anything in there bigger than

the parody drive so I figured i' I just

get a massive one so I can basically put

whatever I want in here and actually be

able to utilize that storage space

should I do a

spread that's how networking nerds Flex

there it is pcie you got to take out

these stupid rubber things you can tell

I actually use it it's dusty there we go

we got these little things out

definitely not putting those back in we

only use uh professional tools here at

Tech cut media pop the screws out and

let's see how easy this

is this doesn't feel good like magnets

I've taken apart the smaller one but

never this

one and there it is let's go and get

this guy out of the way I'm trying to do

this without having to take off the

entire shell I think I'll be able to is

that lined up I can't tell I'm going to

turn on flash so I could see Focus for

me and let's just push let's go for it

hey will this fit back on here that

almost might be too convenient hey I

think I'm going be able to put this back

on you actually hold it uppr right sweet

so it's in that was a pretty simple

installation of course you are going to

need pcie to actually do this so if

you're using a Nas you have to make sure

that it has that or even better if you

diyi it make your own kind of NZ chances

are you'll have a place to put this and

we'll be able to actually use both of

the uh it's kind of Wiggly it should be

fine with this we'll be able to use both

of the chips which is nice which I

honestly probably don't even need but I

don't like to have Hardware that doesn't

at least have the potential to be

utilized and then in the future if

something like image for example starts

to support this I can dedicate one to

frigate and the other to another

application I can actually use it I know

there's other tools other AI softwares

that you can use with this so just to be

able to separate both of them and have

that option is

cool so I'm going to go ahead and put

these screws back in through the hard

drivve Z put it back in the rack and uh

see if we can detect them all right here

we are we have our unrate server running

and supposedly all we need to do is go

over here to apps and search for coral

and right here the coral accelerator

module drivers let's open that up if you

are not on unrated and you're on some

other system these are really easy to

install I'll leave a link down below

with the packages that you'll need

whether that be in a bluetoo system so

on so forth so let's just click install

and see if this is just going to work

it's just a plugin versus a Docker

container which is cool with me let's go

ahead and hit done and now where is this

going to be let's find it there it is

coral driver let's open that up and look

at that we have coral tp1 and coral

tpu2 both alive and well they seem to be

working just fine so now we're going to

install uh frate freegate whatever and I

know there's some things we need to do I

might have the config already in here uh

I have all my configs with an app data

here this right here is a plugin that

allows you to kind of browse your file

system and we have frate which is cool

uh I'm going to go ahead and get rid of

a lot of this real quick cuz this is

basically a fresh install but I am going

to be keeping the config yml so if

you're doing this for the first time

you're going to need to create the

folder and the config file so I'm going

to delete those and let's take a look at

this here we are this is a older

configuration file I've been building a

GitHub repository with or I've been

slowly building it as I do things with

all of my uh home lab stuff so if I head

over here and go over to home lab you

can see surveillance here we have

freegate and this right here is my new

config so I am going to just grab this

entire configuration here and drop it on

in now there is one thing we are going

to need to change and that is up here

where says detect you can see I was only

using one Coral at the moment or at that

specific moment again we see both of

them cooking so we can't add the second

one so I'm just going to copy this paste

this in this is going to be Coral 2 and

I'm assume this is one let's actually

check that so let's open our terminal

here within unraid and we are just going

to run this command to see if they're

showing up we can see both of them I'm

pretty sure that's the same command that

the uh guey for underr is running to

show them and let's list these devices

here and we have zero and one so that

should work and obviously we'll be able

to see that when we actually uh launch

everything so that's said additionally

we have mqtt this is needed if you're

going to link it up to Home assistant I

will change this later to get it

connected properly since now it's going

to be on a separate machine I can't just

use this uh host name but let's go ahead

and save this go back into our app data

FR it delete this one and then let's

upload our new one and there we go so

now let's head over to apps and actually

search for this uh free G and give it an

install Okay within the setup you can

see kind of the instructions some of the

things I talked about which is the valid

configuration we're going to set this to

uh privileged so it can access the coral

tpus and kind of skim through this here

media path we are going to change this

this is going to be under data NVR

there's nothing in there right now which

is good so user data n VR beautiful

we're going to keep all the ports the

same I'll change this password later and

here is the coral mapping so like we saw

in the terminal we're going to use this

so the first one is going to be uh Apex

Z and we're probably going to need to

add another device so let's go down here

add a device iice be tpu2 and drop this

in make it number one add hopefully that

works we're going to keep our Intel in

there since it does have a actual Intel

GPU you are able to use multiples as it

says up here actually optionally but

highly recommend that you can use

multiple devices to perform object

protection such as the coral accelerator

and a Intel GPU so we are going to get

rid of this Nvidia garbage remove remove

I'm just going to keep this as is for

now I do need to do a little bit more

research in that let's do it let's run

it hit apply and just like with

installing the uh the drivers initially

like we did if you're not doing this on

UNR it's a simple uh Docker Docker

compose file that you can use and

there'll be a link to frate over here in

all their documentation so if I go under

like getting started it goes over just

about everything that you are going to

want to know and use and configuring

this is really dependent on the hardware

that you want to use and honestly it

does take a whole lot of playing around

with the configuration to actually get

everything to work properly so do note

that all right so I did have a little

more work to do I had to change these uh

IP addresses for the rtsp path to

properly match with the actual local IP

address of this container um if I go

ahead and start it up here let's go into

our logs now see if we still have

issues regardless of those issues right

here user and password this is my

current administrator password so let's

grab it head over here and paste it in

log in and there we go and we do have

some fees so part of it is working oh

there's been some updates this looks

really pretty let's head over to our

settings real quick and go to system

metrics here we have our GPU usage we

have our interface speeds and

temperatures and you can see we have

Coral 1 and coral to so it is utilizing

both of the tpus now which is great as

well as the gpus so we have multiple

things working you see it's still trying

to figure out what's going on there I

have a feeling that's the problem with

the real link and VR that thing is up

pain pain in the ass but both of them

are showing up which was the whole point

of everything that I'm doing so as of

now I'm going to call this a success as

you can see we still have a lot of

errors in the logs uh this is not a uh

fate or frate configuration tutorial so

knowing I got these two working I just

have probably some slight configuration

changes to make to uh get all this up

and running but we're now utilizing both

of those uh chips which is freaking

awesome but yeah that's how you get

these uh Coral tpus working with fry I

do hope you enjoyed this video hope you

have a fantastic day and goodbye