• expired

Google Coral USB Accelerator $110.99 Delivered @ RS Components

801

I've been keeping an eye out for these for my Home Assistant / Frigate setup, but they've been at least $150 even on RS previously. Happened to check and see they are down to $110.99 with free delivery so bought one!

The Coral USB accelerator by Google, allows for an edge TPU coprocessor to be added to any system, enabling high-speed machine learning inferencing. It includes a USB-C socket to connect to a host computer to perform accelerated ML inferencing. The on-board Edge TPU is a small ASIC designed by Google that accelerates TensorFlow Lite models in a power efficient manner: it's capable of performing 4 trillion operations per second (4 TOPS), using 2 watts of power—that's 2 TOPS per watt.

Edit for GST!

Related Stores

RS Components
RS Components

closed Comments

  • +9

    C'mon this isn't a deal it's way above RRP, I have a few of these, normally purchase via SeedStudio for around $90 AUD, although they seem out of stock today.
    If you have a spare M.2 slot, they are ~$25 USD, good use of a spare WiFi M.2 slot and lower latency than USB.
    They come in M.2 A+E and B+M so ensure you order the correct product.

    These are all slower than a typical gpu, but I like them for Image AI as they consume so little power, and give off so little heat.

    • OOS but agree. There is no bargain here.

    • +5

      This is the first time I've seen it in stock and below $150 since I've been looking (~6 months) so thought it was worth posting for just above RRP for those who are keen. Amazon / eBay are all still considerably more.

      • Then you are looking in the wrong place, but agree this is a decent price, especially today as everyone else appears out of stock or asking stupid money.

        Go via https://www.coral.ai/ to see the official resellers, these are regularly in-stock, I have a few all purchased at different times for various builds. I mainly use these when I setup NVRs for people using code project ai paired with BlueIris.

        • RS Components is on that list - I'd love to see an actual link to one in stock and cheaper.

          • @jamhos: Please re-read the first line above where I agreed this is a decent price, especially today.
            As I noted, I regularly purchase these from seedstudios, they pop in and out of stock all the time.

    • Give us a link to where we can buy it cheaper son.

      SeedStudio shows as out of stock for all forms of the Coral for me.

      • I noted that these come in and out of stock all the time, it's a good "community notice" but really, still selling above RRP to be classified as a deal?
        If I needed one TODAY then sure I would consider RS OnLine and doubt there is a better price TODAY, but would not recommend the USB version anyway.

        Link to the M2 version below, which also comes in DUAL TPU for less than the USB version

        https://au.rs-online.com/web/p/communication-wireless-develo…

        • +2

          I got a Lenovo micro PC which doesn't play nice with M2 so gotta go USB sadly.

          I've been looking on and off for a while and this is the cheapest I've seen delivered and in stock.

          RRP means nothing if no one has had them in stock for that price for a very long time.

        • Dual TPU requires a compatible PCIe slot but many are just sata.

    • any link for a PCIe version ?

    • Still no stock at seed studio. In any case they're US 59.99 so you need to add a bit more to get free shipping.

      Also you are taking a bit of a gamble with m2 versions if you are using with cheap used USFF PCs as some manufacturers (eg Dell) can restrict what you can plug into it.

  • +1

    Can it run Crysis?

    • Or Crysis to run?

    • +3

      No mate, unfortunately it cant. I would have gotten one otherwise.

  • +3

    I believe the pcie one is actually faster because of the higher bandwidth

    And way cheaper

    • Also more reliable. Downside though, is depending on your use case, you need to add your own cooling to the PCIE ones and power requirements may be unsuitable for some motherboards.
      Personally though, all my use cases like CodeProject AI throttle the Coral so no need for external cooling and many DIY options to cool.

  • +3

    Bought one to plug into USB slot. Hope it's good at doing USB things

    • +2

      Don’t worry, I’m sure you won’t remember what you were going to do with it by the time it arrives.

  • +3

    Wow … Half way down reading comments and thinking why the heck would people buy this for their FRIDGE … Until I realised , I am also too stupid for this post and should put my glasses on

  • This is a cool bit of kit, is frigate the open source NVR of choice?
    What cameras are you all running with it?

    • +1

      I'm running Reolink. best bang for buck cameras out there

      • Thanks, I forgot about that brand, I'll check them out. Since they're ip cameras I guess no special reolink software is required?

        • +1

          Some people have had issues with Reolink and frigate. Others haven't. I haven't. There are Reolink specific config settings in the documentation. read up

  • +1

    I'm a fair bit of a computer nerd and still have, on face value, no idea what this is useful for…

    Feeling a little out of the loop lol.

    • +2

      They're great for machine learning. Lots of the more tech savvy likely use them for object detection in home network security system

      • Ahhh thanks, that's an excellent short explanation for me.

        Seems very niche.

  • Weird Google don't even sell this on Google Store.

  • +1

    So… it doesn't actually accelerate USB speed??

  • +3

    I bought the m.2 version when it was cheap, waited 14 months to get it but dropped my frigate detection power requirements from ~150w on CPU to ~4w with the coral

    • Diff form factors have varying compatibility with hypervisors. A bit confused as I read only the USB can be passed through in esxi. Are you using it on bare metal or another hypervisor?

      • +1

        I have Frigate running as a docker on TrueNAS Scale, took a bit of mucking around to get it working

  • Hey guys, I've got a home assistant set up using a mini pc I bought here. It's a cheap 6th or so Gen i5 or something that I got for like $100 delivered, and no gpu or anything.

    I've heard of frigate and how amazing it is but never thought I'd be able to get it because of how much processing power was needed so haven't done any further research into this. Would this let me use frigate as is just by plugging into my mini pc? I don't need to get anything else like a gpu or anything to make it work? If so, that's amazing!

    And with frigate, can you use any smart camera you want and connect it up and have local control within HA, or are there limitations? I have a eufy cloud based one that was gifted to me that would be awesome to control locally!

    Thanks in advance!

    • Yep that’s all I’ve got, it makes it so CPU doesn’t have to do too much at all. Won’t need a GPU.

      Re: cameras, you can use pretty much anything with an RTSP stream. I’m using a Eufy cam at the moment but it’s a wired one - wouldn’t want to use battery as they don’t really do 24/7 streams. Of course there are better ones but it does the basics.

      • That sounds pretty perfect - I've got a eufy wired one too so it sounds pretty great! Especially since I don't need anything else to make it work! Thanks for that!

    • +1

      Best if you use Intel QuickSync on your CPU for the video encoding\decoding.
      The coral comes into play for Object Recognition, ie: person, animal detection.
      Frigate is great, although my go-to preference is BlueIris (NVR software) and CodeProJectAI(Object Recognition Software), LPR ect…

  • Will this make my Google Home run Crysis?

  • +1

    was this the MacGuffin from the movie Sneakers?

  • -1

    I do all my ML on AWS for free. :D

    • This isn’t for ML

    • I use Amazon Polly for free, but generated a few lines using their non free voices. Months later they send me a bill for 6 cents. Client didn't even end up using the robot voice, so I wasted 6 cents for nothing.

  • I would love to receive this as a Christmas present. Impossible to happen since my families won't even know what this is.

  • +2

    If I buy this will I become smart too like OP?

  • I'm just here to read the entertaining comments XD

  • +1

    Hilarious comments, but I am surprised that no-one has actually mentioned the real reason for buying this is to help detect when a red footed booby enters your parent's yard.

  • I got one from seeedstudio for about $100 delivered. They are really cool. I want to make a device to detect and scare off pigeons but leave natives. So far I hav the device set up and running on an RPi 400. If I upload any picture of a pigeon it detects it with about 90% confidence.Got a lot more steps to go before I have an operational pigeon scarer but pretty impressed with the device

  • BTW took hours to set up on win 11 and about 15min on an RPi 400

    • Hmm.. really unsure why, I deploy lots of these in W11 it's really a 1-click install, assuming you already have "Microsoft Visual C++ 2019 redistributable" installed.

      Just download and extract the zip and run install.bat, never had to do anything else and I've done about 30 of these so far.

      https://coral.ai/docs/accelerator/get-started/#runtime-on-wi…

  • Well, there are at least 60 people who know what OP is talking about

    • I'm still not sure what it does. Some kind of specialised low energy processor for machine learning?

      • Basically yes - it's got special hardware for Google's TensorFlow libraries used to create ML models. Think of it like a specialised graphics card

      • Yep , great for object detection for your home NVR that's running on a pi, lower power for the pi and coral

  • The picture makes it look postage stamp sized, but it's actually a lot bigger than that in other photos.

  • This is one of the best comments to help understand what this cool thingy can do

    "I want to make a device to detect and scare off pigeons but leave natives."

    And I thought this was a joke, but it's actually valid

    "the real reason for buying this is to help detect when a red footed booby enters your parent's yard."

    Thanks for all the comments everyone. I've actually learned quite a bit about a subject I didn't know existed. 🤣

  • Ive been using the coral edge tpu m2 which has been working flawsley with frigate. Also cheaper than the usb option.

    https://au.mouser.com/ProductDetail/Coral/G650-04527-01?qs=X….

  • Could this (the pci/ m2 version obviously) be used with a VMWare ESXi host, possible passed through to a specific VM?

  • Don't understand what it is. I thought it was a wireless charger

  • -1

    Those explanations with words of more than fifteen syllables are all helpful and surprisingly accurate. But simply this is just a flux capacitor that in 1985 took up half a car but that now thanks to Moore and his Law takes up less than half a car. It's used to get part of your life back … to take you back in time to just before you clicked on this OB link and started reading this crap.

  • Worthwhile noting that it seems MS will set a 45-50 TOPS requirement for Win12, and therefore there is a good likelihood that plug in boards with 10x the power will come out for those looking to bolster their desktops/laptops for the new OS over the coming year.

    And for those looking around bewildered - these are inference accelerators for helping with machine learning, not training. People use them mainly for applying ML to video feeds on home NAS so they can detect and classify objects to yield smart security cameras (eg not triggered by the dog).

  • Interesting and cool at first sight, however only makes sense if you need AI inference on a low power device.
    Otherwise any almost decent GPU which might be already in your PC/Laptop will deliver more TOPS at no additional hardware cost. If you have an NVIDIA GPU then any software will support it.
    Btw, NVIDIA RTX 4090 does about 660 TOPS (INT8), so just assuming 330W Power consumption (for easy math), you also get ~ @ TOPS/W
    I dont have the number for the other smaller GPUs at hand but should be similar .. and always faster than this small USB device.

    If you just want it a bit faster, and not too big, this is for you :
    https://developer.nvidia.com/embedded/jetson-developer-kits

    P.S.: If you do not understand what I am saying, don't buy, because you wont need it or be able to put it to work.

    • True, I use this for NVRs get around 130ms for image recognition vs 30ms with a 3060ti. However with the coral I can build with passive cooling only and in a device running 24x7 the power saving is real. But I use the M.2 slots, typically the wifi slot which is spare in most of my NVR builds as I use wired only.

    • works great with NAS that can do docker + Frigate. my CPU on NAS <20% which is mpeg decoding. I have about 5 reolink duo2 on frigate, pretty smooth going

  • i had bought one recently and recieved within a week. recommend rs-shop. I thought this was RRP and didnt think of posting but looking at the votes, looks like its worth it.

  • Mine has not arrived yet since ordering on the 20th, and they just charged my card again. Sigh, time to call them

    • Same! Still no tracking number

  • +1

    Mine arrived today :)

Login or Join to leave a comment