• State Resolved
  • Locked Locked
  • Replies 7 replies
  • Subscribers 99 subscribers
  • Views 422 views
  • Users 0 members are here
Support feedback
Options
Options
Related

This thread has been locked.

If you have a related question, please click the "Ask a related question" button in the top right corner. The newly created question will be automatically linked to this question.

AM625: How can we implement AI solution

Ryuuichi machida
Guru 12620 points
Part Number: AM625
Other Parts Discussed in Thread: SK-AM62, TDA4VM

Hello,

According to following Git repo, TI AI solution is supported on AM62.
https://github.com/TexasInstruments/edgeai-tidl-tools

However, even though I checked product page of AM62x and SK-AM62, there is no information for SDK.
Which SDK should I use to implement AI soultion for AM62x ?
Or, even though there is no device name on supported hardware, can I use same package of TDA4VM(ti-processor-sdk-linux-sk-tda4vm-etcher-image.zip) ?

BR,

  • 0
    Guru 12620 points

    Hello,

    Could you please give me about feedback ?

    BR,

  • 0 in reply to Ryuuichi machida
    TI__Guru* 83745 points

    Machida-san

    The expert who has been assigned to this post is out of office this week. Please expect delays. 

    Can you explain further on what are your application and performance requirements for AI solution. 

  • 0 in reply to Mukul Bhatnagar
    Guru 12620 points

    Hello,

    >Can you explain further on what are your application and performance requirements for AI solution. 
    We can not decide exact performance. However, since it seems that TI only show two kinds(TDA4 and AM62) of devices which can be used for AI.
    I understand that there are a lot of informations for AI but cost of this device is expensive.
    Therefore, I would like to understand low cost AM62x device AI solution.
    If you have any other devices which you recommend to use AI inference, please let me know.

    BR,

  • 0
    TI__Genius 15056 points

    Machida-san,

    AM62x does not have accelerator hardware, but can still be used as an Arm-only AI solution (4x A53s at 1.4GHz). The Tensorflow-lite and ONNX runtimes are available for use in the Linux SDK for AM62 -- this is what you should use for developing your solution. You may use C/CPP APIs or the python libraries for these two runtimes. The edgeai-tidl-tools is using these two runtimes on AM62.

    You should not use the TDA4VM image on this device, please only use the AM62 images/SDK.

    Cheers,

    Reese

  • 0 in reply to Reese Grimsley
    Guru 12620 points

    Hello Reese-san,

    Thank you for your reply.
    >please only use the AM62 images/SDK.
    Where can I get this image ?
    I understand that Linux SDK should be downloaded from below URL.
    www.ti.com/.../SK-AM62

    However, there is no SD image file such as TDA4VM SK.(Or rather, edge ai SDK is only available SD card .)
    Here is SD image for TDA4VM SK.
    www.ti.com/.../SK-TDA4VM
    And also here is Linux SDK for TDA4VM.
    www.ti.com/.../TDA4VM
    (I think that above image file is not same one as creating by Liunx SDK...)

    In case of TDA4VM-SK, I can use above image to write SD card directly.
    However, in case of AM62x-SK, there is no image file which can be written to SD card directly.

    Is my understanding when user want to use AI on AM62x-SK, should user implement Linux SDK to Linux machine such as VMware and create SD image on it correct ?

    BR,

  • +1 in reply to Ryuuichi machida
    TI__Genius 15056 points

    Hello Machida-san

    Yes, you can get the SDK from that link or from here:

    https://www.ti.com/tool/download/PROCESSOR-SDK-LINUX-AM62X

    There is an SD card image on this page: the .WIC.XZ file. You do not need to setup the Linux SDK on a host machine unless you need to make specific changes to the image or build applications for AM62x from the host machine.

    The TDA4VM has a version that explicitly says "Edge AI SDK", but this is mainly because it comes with additional out of box demos for Edge AI, which it has an accelerator for. The default Linux SDK for AM62x includes the runtimes needed for developing Edge AI applications, which will run on the Arm cores.

  • 0 in reply to Reese Grimsley
    Guru 12620 points

    Hello Reese-san,

    Thank you for your reply.
    Understood.

    BR,