$70 UP AI Core mini PCIe Card Features Intel Movidius Myriad 2, Supports Tensorflow and Caffe Frameworks

AAEON’s Up Board has given us some affordable Intel development boards over the years with products such as the Cherry Trail based UP Board, or Apollo Lake powered UP Squared board among others, that are competitively priced against equivalent Arm development boards.

The company has now launched a new UP AI Edge family, which will include hardware based on Intel Altera FPGA or Intel Movidius VPU (Vision Processing Unit). Their first product is based on the latter, more exactly Movidius 2 2450 VPU, and instead of being a standalone board, UP AI Core is a mini PCIe card that can fit into any 64-bit Intel board or computer.

UP AI Core card specifications:

  • SoC – Intel Movidius Myriad 2 2450 VPU
  • System Memory – 512MB DDR SDRAM
  • Mini PCIe edge connector
  • Dimensions – 51 x 30 mm

Host computer/board requirements

  • System Memory – 1GB RAM or more
  • Storage – 4GB of free storage
  • Free mini PCIe slot
  • x86_64 computer running Ubuntu 16.04

The card supports TensorFlow and Caffe frameworks for now. It looks similar in function to Movidius Neural Compute stick, but instead of an USB 3.0 interface, a mini PCIe is used. Since the stick works with Raspberry Pi 3 board, it’s quite possible the AI Core also work on Arm boards with an mPCIe slot. While the company does not explicit mentions the interfaces used over the mPCIe connector, the card is compatible with UP Squared board, so data is likely going over PCIe instead of just USB.

The company explains the the board enables enables rapid prototyping, validation and field deployment of Deep Neural Network (DNN) inference applications at the edge without a connection to the cloud. Intel Movidius Neural Compute SDK allows developers to profile, tune, and deploy Convolutional Neural Network (CNN).

AAEON is taking pre-order for UP AI Core for $69.99 with shipping scheduled for the end of April.

Share this:
FacebookTwitterHacker NewsSlashdotRedditLinkedInPinterestFlipboardMeWeLineEmailShare

Support CNX Software! Donate via cryptocurrencies, become a Patron on Patreon, or purchase goods on Amazon or Aliexpress

ROCK 5 ITX RK3588 mini-ITX motherboard

12 Replies to “$70 UP AI Core mini PCIe Card Features Intel Movidius Myriad 2, Supports Tensorflow and Caffe Frameworks”

  1. @cnxsoft: ‘so data is likely going over PCIe instead of just USB’

    If the pictures are correct it can only be PCIe since the relevant pins are routed to somewhere on the PCB while pins 36/38 (USB2) aren’t.

    BTW: I hope you feature the new ‘UP Core Plus’ board soon too and show there especially the passive heatsink so other board vendors might learn how efficient passive cooling looks like (large distance between large heatsink fins): http://i0.wp.com/www.up-board.org/wp-content/uploads/2018/02/upc_stack.png

      1. Since the net plus has a mPCI-e, can I use the Core Plus + Net Plus + AI Core together? If this is possible, this could become a good alternative to Jetson TX2.

  2. @tkaiser
    Yes, the pins are PCIe, but both of you missed the Fresco Logic USB 3.x host controller on the board.
    In other words, the card itself is mini PCIe, but the Myriad 2 is still connected via USB through the host controller.
    Odd implementation, but I guess it makes it easier to plug this board into a variety of hardware.

  3. @TLS
    Ah, interesting. Indeed looks like a FL1100 used here. Just had a quick look at the specs and the various Atoms that are available on UP Core Plus seem to have 3 USB3 connections and 6 PCIe 2.x lanes. And only by using either ‘Net Plus’ or ‘AI Plus’ Expansion boards the mPCIe/mSATA connector (seems to be switchable) is available so most probably they ran out of USB3 ports and had to use a PCIe lane then?

  4. @tkaiser
    Well, for one, only M.2 supports USB 3.x, whereas mini PCIe only supports USB 2.0, so this is one limitation based on the choice of connector on this board.
    Then by going down this route, there is obviously a much wider ecosystem that you can plug it into, especially when it comes to more industrial applications. A simple PCIe to mini PCIe adapter would allow you to use it with any old desktop as well. This would be harder to “borrow” than a USB dongle and also harder to break, as you can’t bump into this by accident.

    The PCB appears to have space for a micro USB-B 3.0 on the opposite end, but it’s not fitted to the PCB and may require additional components to function. Maybe we’ll see an external version in the future that connects over USB 3.0.

    @cnxsoft there’s a small typo, 512MD instead of 512MB. Seems like a lot of integrated RAM in such a small chip though and the official spec states 2MB.

  5. @TLS
    Thinking a little bit more about use cases the mPCIe/mSATA connector starts to make a lot of sense now even if for this use case a PCIe USB3 controller had to be ‘wasted’:

    * the Myriad 2 most probably performs better with an USB3 SuperSpeed connection than USB2
    * the mPCIe connector is limited to USB2, PCIe or mSATA (no USB3 here)
    * AAEON most probably wants to provide their users with the most flexible miniature expansion port possible

    Though M.2 key B could’ve been also used IIRC (PCIe x2, SATA, USB2 and USB3)?

  6. @tkaiser
    My comment saying almost exactly that was held for moderation…

    M.2 key B is usually limited to SATA drives only, even though technically it’s meant to support additional interfaces.
    I have not seen any implementations where USB 3.0 is actually connected to an M.2 key B slot.

  7. @tcmichals
    From that forum post, I understand the camera needs to be tuned by Intel and the camera manufacturer, so only the camera sold by UP Board will be supported. While they are working on it, no such camera module existing yet on their shop, so the only solution for now is to use a USB 3.0 camera.

Leave a Reply

Your email address will not be published. Required fields are marked *

Boardcon Rockchip and Allwinner SoM and SBC products
Boardcon Rockchip and Allwinner SoM and SBC products