Hey Raspberry! Can you send me your latest board (Raspberry Pi 5)? Im working on a startup! (also an engineer who can work on ML to detect baby's cries!) XD
100% chance it was your neighbor who narc'd on you. Pick the lock on his apartment and fill it with 1000 live crickets..call building security and complain your neighbor is doing something weird with insects.
not a critique just an fyi - your video has a lot of low frequency audio that is a bit distracting (mostly during typing). most folks don't have a subwoofer that goes down to
I used the Lenovo ThinkCentre 720Q. A refurb was less than an RPi. It's got an Intel I5 so way more power, comes with NVME 256, 8 gigs, ready to rock. Install Ubuntu. It's got a proprietary PCIx8 for which you get a riser adapter for $20, from there to the M.2 coral PCIx2 to 1 adapter to the M.2 coral edge TPU. You have to remove the external eSATA adapter from motherboard so card fits. Used a jigsaw to cut a hole for chassis fan for active cooling. Worked right away and never gets above 45 deg, under sustained stress tests. The M.2 to PCI coral adapter has a 12v via for fan leads. The whole setup has a nice enclosed profile only slightly larger than RPi jankery.
Have you tried installing the M.2 coral on the wifi slot? I got a hp prodesk ssmf (i7 8th T series processor, 8 GB RAM). The only option I see is the coral accelerator (usb device)
@@Paw_Low That probably wont work. There are different M.2 protocols. You need the "E-key" type. Even so, Although the M.2 Specification (section 5.1.2) declares E-key sockets provide two instances of PCIe x1, most manufacturers provide only one. To use both Edge TPUs, be sure your socket connects both instances to the host.
I hope you can upload new video explaining about the frigate after installation, maybe u can implement some models so you show how the computer vision works. Thanks man
I bought a dual edge coral tpu and I follow every step carefully, changing config.txt, kernel and headers. But only apex_0 is detected. what is missing?
why oh why ? , why a dual TPU ? , my frigate set up is 6 cameras using a usb TPU and a ryzen 5900g running proxmox . With a raspberry PI 5 the CPU and GPU would be the limiting factor before you would need a dual TPU , every camera needs to transcode the stream and remove the audio , frigate works by looking for movement and scaling parts down and passing the frame to the TPU to analyze , so most of the work is done with the CPU / GPU the only time the TPU comes into play is with movement to check is it a cat , dog , bird or superman etc .
So, you're saying that the PI 5 has a CPU/GPU that is too limited for this task (for managing multiple cameras) and therefore the association with the Hailo-8L chip is not appropriate? It's a bad mix of computing capacity (with a bottleneck because of the PI 5)? A Hailo-8L M.2 chip would be more suitable on a NUC with a better CPU/GPU? In fact I'm looking for the hardware needed to manage HAOS+Frigate recognition with more than 4 cameras (including 4k) + basic home automations. I looked at second-hand NUCs, but it's quite complicated to know from which generation and which configuration it becomes possible to run all that (Coral USB looks like it's outdated, I would have preferred Hailo-8L).
@@xaverty No , what I'm saying that a single tpu will process 100 fps or 20 cameras at 5 fps , so why the dual tpu , unless you have over 20 camera / doing more than 100 fps why a dual. so if you have 20 plus cameras the cpu/gpu will give out before the single tpu. With my set up I've 2x4k , 2x8mp , 2x4mp camera and using 1.5% tpu max and 5% cpu/gpu , so WHY A DUAL TPU ?.
Seems like alot of lag in that processing pipeline. I am getting several hundred frames per second from the dual TPU's baseline. For full object segmentation using several models I see sub 20ms per frame with one TPU.
@@peteradshead2383 @Sven_Dongle also running the Pi over wifi on a TPLink router. Lot's of hmmmmm going on with that. I would have put MQTT in a container as well. Just simpler to manage when an update comes out.
Always. You ALWays wanted that. smh . No. No you didn't always want that. You didn't think about it, so you didn't want it. Does anyone hear themselves think anymore?
man what's up with that audio? your voice sounded like it was ai or something. maybe too much compression? will have to nope out on finishing watching this. sorry, looked interesting.
DON'T get evicted...
Hey Raspberry! Can you send me your latest board (Raspberry Pi 5)? Im working on a startup! (also an engineer who can work on ML to detect baby's cries!) XD
100% chance it was your neighbor who narc'd on you. Pick the lock on his apartment and fill it with 1000 live crickets..call building security and complain your neighbor is doing something weird with insects.
lmao
🤔😁😆😑😆😅😂🤣😭..🥳
This made me lol
So what's the AI part of this project?😊 I thought it was going to recognize the person that let you know?
yeah the lies in the title get absurd
Thanks for the setup, I didn't see any AI going on but I guess bounding boxes are a tickbox or something
not a critique just an fyi - your video has a lot of low frequency audio that is a bit distracting (mostly during typing). most folks don't have a subwoofer that goes down to
Respect to you for figuring all that out...
I used the Lenovo ThinkCentre 720Q. A refurb was less than an RPi. It's got an Intel I5 so way more power, comes with NVME 256, 8 gigs, ready to rock. Install Ubuntu. It's got a proprietary PCIx8 for which you get a riser adapter for $20, from there to the M.2 coral PCIx2 to 1 adapter to the M.2 coral edge TPU. You have to remove the external eSATA adapter from motherboard so card fits. Used a jigsaw to cut a hole for chassis fan for active cooling. Worked right away and never gets above 45 deg, under sustained stress tests. The M.2 to PCI coral adapter has a 12v via for fan leads. The whole setup has a nice enclosed profile only slightly larger than RPi jankery.
Have you tried installing the M.2 coral on the wifi slot? I got a hp prodesk ssmf (i7 8th T series processor, 8 GB RAM). The only option I see is the coral accelerator (usb device)
@@Paw_Low That probably wont work. There are different M.2 protocols. You need the "E-key" type. Even so, Although the M.2 Specification (section 5.1.2) declares E-key sockets provide two instances of PCIe x1, most manufacturers provide only one. To use both Edge TPUs, be sure your socket connects both instances to the host.
The fun version is the asus carrier card that can run 8 of them in on PCIEx16 slot.
How can you use both TPU if you need 2 separate lane to use it? Pi5 has only 1 lane so you'll always use only one of the processors.
Is there a way to make a large bank of these? Some way to multiplex the ports? I would like to make a cluster with 256 of these puppies
Aren't coral tpus out of stock?
I hope you can upload new video explaining about the frigate after installation, maybe u can implement some models so you show how the computer vision works. Thanks man
Surely you can use a raspberry pI now with the ai board?
next project: ai robodialing my neighbor until he goes insane
I bought a dual edge coral tpu and I follow every step carefully, changing config.txt, kernel and headers. But only apex_0 is detected. what is missing?
What all hardware are you using?
@@DataSlayerMedia sorry..I am using wrong pineboard hat. Thanks for your response
my Pi5 has a M.2 HAT for my NVME/SSD - you are not using an SSD for your Pi5? I need two M2. slots for what you are showing, any recommendations?
Thanks, great project.
Data Slayer, You're so talented! I had to hit the like button!
Where did you get the heatsink you used?
Your neighbor seems like a super fun guy to be around 🙄
It's entirely reasonable not wanting to be videoed coming in and out of your home? Not everyone wants to live in China, including most of the Chinese
why oh why ? , why a dual TPU ? , my frigate set up is 6 cameras using a usb TPU and a ryzen 5900g running proxmox .
With a raspberry PI 5 the CPU and GPU would be the limiting factor before you would need a dual TPU , every camera needs to transcode the stream and remove the audio , frigate works by looking for movement and scaling parts down and passing the frame to the TPU to analyze , so most of the work is done with the CPU / GPU the only time the TPU comes into play is with movement to check is it a cat , dog , bird or superman etc .
So, you're saying that the PI 5 has a CPU/GPU that is too limited for this task (for managing multiple cameras) and therefore the association with the Hailo-8L chip is not appropriate? It's a bad mix of computing capacity (with a bottleneck because of the PI 5)? A Hailo-8L M.2 chip would be more suitable on a NUC with a better CPU/GPU? In fact I'm looking for the hardware needed to manage HAOS+Frigate recognition with more than 4 cameras (including 4k) + basic home automations. I looked at second-hand NUCs, but it's quite complicated to know from which generation and which configuration it becomes possible to run all that (Coral USB looks like it's outdated, I would have preferred Hailo-8L).
@@xaverty No , what I'm saying that a single tpu will process 100 fps or 20 cameras at 5 fps , so why the dual tpu , unless you have over 20 camera / doing more than 100 fps why a dual.
so if you have 20 plus cameras the cpu/gpu will give out before the single tpu.
With my set up I've 2x4k , 2x8mp , 2x4mp camera and using 1.5% tpu max and 5% cpu/gpu , so WHY A DUAL TPU ?.
I found a real gem I guess. Perfect channel, gratz
Where are the TPUs actually used? Disappointing demo
Can this work on Rasp Pi 4?
will this work on thunderbolt 4 to m.2 pcie?
Great work ...thanks
15:24 - that chicken store name :D
Thx for sharing...
I like the subject but why I never know what are you talking about?
You Neighbor isn’t happening lol😂him got recorded
nice vid corndog! i like that.
Seems like alot of lag in that processing pipeline. I am getting several hundred frames per second from the dual TPU's baseline. For full object segmentation using several models I see sub 20ms per frame with one TPU.
A raspberry pi 5 and no hardware acceleration , no the best set up .
@@peteradshead2383 @Sven_Dongle also running the Pi over wifi on a TPLink router. Lot's of hmmmmm going on with that.
I would have put MQTT in a container as well. Just simpler to manage when an update comes out.
fyi you doxxed yourself at 15:20
bro bring new video on miner
eye of sauron -- lol
Subscribed af
Always. You ALWays wanted that. smh . No. No you didn't always want that. You didn't think about it, so you didn't want it. Does anyone hear themselves think anymore?
man what's up with that audio? your voice sounded like it was ai or something. maybe too much compression? will have to nope out on finishing watching this. sorry, looked interesting.