• State
  • Locked Locked
  • Replies 7 replies
  • Subscribers 103 subscribers
  • Views 318 views
  • Users 0 members are here
Support feedback
Options
Options
Similar topics

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.

AM69: C7x Load Tests for Thermal Profiling

Parth Pancholi
Intellectual 290 points
Part Number: AM69

Tool/software:

Hello TI Experts,

We have a custom HW based on AM69. Would like to validate the thermal profiling of the HW with a full load on CPUs and all 4 C7x (32 TOPS) + MMA accelerators available on SOC.

Seems like below demo/utility from the TI RTOS SDK's Vision Apps is relevant for our use-case, however as far as I understand, Linux + RTOS SDK is not supported on AM69.

https://software-dl.ti.com/jacinto7/esd/processor-sdk-rtos-j784s4/09_02_00_05/exports/docs/vision_apps/docs/user_guide/group_apps_utilities_app_thermal_load_cpu.html

How can we enable these RTOS demos on AM69 device ?

Just to note that with `ti-rtos-firmware` recipe for C7x firmware, we are able to see all 4 C7x DSP cores (32 TOPS) up and running on our platform.

Would need help/support for the demos/utilities which can simulate load on each C7x + MMA.

root@aquila-am69-02242602:~# cat /sys/class/remoteproc/remoteproc*/name
64800000.dsp
65800000.dsp
5e00000.r5f
66800000.dsp
5f00000.r5f
67800000.dsp
5900000.r5f
5a00000.r5f
root@aquila-am69-02242602:~# cat /sys/class/remoteproc/remoteproc*/state
running
running
running
running
running
running
running
running

Thank you.

Regards,

Parth P

  • 0
    TI__Genius 10170 points

    Hi Parth,

    We don't support running Vision Apps on the SK (AM variants) of the boards. However, if you are looking for a similar print out as that demo, you can utilize the perf-stat tool that is part of edgeai-gst-apps. More info here. Let me know if you have further questions. 

    Best,

    Asha

  • 0 in reply to Asha Bhandarkar
    Intellectual 290 points

    Hello ,

    Thank you for response.

    If I understood you correctly, perf-stat which you suggested would only help print the current load on the C7x cores which is just the small part/monitor gate for the actual tests which I would like to do.

    As I mentioned, I would like to test all C7x cores with some load so we might see increase in power/current consumption which may lead to overall increase in SOC/HW temperature.

    Could you please suggest some tools/ways by which we can simulate/load C7x cores on AM69 ?

    Also, It would be a huge help if you could share document describing the maximum power/current consumption by these C7x cores present on AM69 SOCs when those are fully utilized by some use-case or dummy load.

    Appreciate your support on this.

    Regards,

    Parth P

  • 0 in reply to Parth Pancholi
    TI__Expert 8235 points

    Hi Parth,

    The best way to exercise the C7x's to near-max thermal performance would be to:

    1) Run the Edge AI Demos (not available on SK..)

    2) Edit the configuration .yaml files in the 'configs/linux/' directory. One technique we have had some success with was to reconfigure model0 path under 'models' to point to the ONR-SS-8610-deeplabv3lite-mobv2-ade20k32-512x512 script instead of the default TFL script that is populated. This increased C7x utilization to 98/99% (confirmed by perf_stat) with a J7 variant.

    Regards,

    Colin 

  • 0 in reply to Colin Callaghan
    Intellectual 290 points

    Dear ,

    Thank you for response.

    I manged to make Edge AI Demos working on our platform and tried your below suggestion.

    Colin Callaghan said:
    2) Edit the configuration .yaml files in the 'configs/linux/' directory. One technique we have had some success with was to reconfigure model0 path under 'models' to point to the ONR-SS-8610-deeplabv3lite-mobv2-ade20k32-512x512 script instead of the default TFL script that is populated. This increased C7x utilization to 98/99% (confirmed by perf_stat) with a J7 variant.

    I have tried modifying couple of config files such as /opt/edgeai-tiovx-apps/configs/linux/image_classification.yaml and /opt/edgeai-tiovx-apps/configs/linux/multi_input_multi_inference.yaml by replacing model0 path to ONR-SS-8610-deeplabv3lite-mobv2-ade20k32-512x512 as you suggested, restarted the system/respective demos and could not see any difference in C7x_[1-4] or any other cores utilization. For eg. all C7x_[1-4] still at ~44% with multi channel demo.

    Am I missing something here ?

    Appreciate your support on this.

    Regards,

    Parth P

  • 0 in reply to Parth Pancholi
    TI__Genius 15080 points

    Hello Parth,

    First of all, what device variant are you using for your project? AM69 or AM69A?

    If it is AM69, ideally you should not be using the C7x cores. Let us know.

    Thanks.

  • 0 in reply to Fabiana Jaimes
    Intellectual 290 points

    Dear ,

    We are using AM69A for our project.

    Just for you to know, I am able to run the EdgeAI demo on my AM69A based HW. In addition, I also see the utilization of the C7x reaching near to ~44% with multi-channel GUI demo packed with EdgeAI demo image.

    Thank you.

    Regards,

    Parth P

  • 0 in reply to Parth Pancholi
    TI__Genius 10170 points

    Hi Parth,

    I do apologize for not getting your question resolved earlier. 

    I've booted the default image for AM69A - there does seem to be a vision_apps folder with some limited support for certain applications/demos. One of these seems to be the load test (see picture below). 

    Can you try and see if you have the same and running the test? It's my understanding this would directly would need and then you can run your various thermal tests as necessary. 

    Best,

    Asha