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.

TDA4VH-Q1: High CPU utilization sporadically

Part Number: TDA4VH-Q1

Hello,

  • SDK: 8.6
  • Board: TI EVM
  • app: 6-camera app
  • we did not change something of R5F(2-0) and R5F(4-0)

We observed that R5F(2-0) and R5F(4-0) sporadically had high CPU utilization, like right after booting or after some time.

And the rates were decreased to low and spiked again and decreased... repeatedly.

We did not see this much with the previous SDKs.

I think this is abnormal.

Best regards,

  • Hi,

    May I know if you are using the multi-cam app present in the SDK 8.6 in your usecase?

    If yes, are there any changes made with respect to this app?

    We did not see this much with the previous SDKs.

    I believe that the previous SDKs were not using MCU4_0 in this usecase. Had you enabled the VPAC on MCU4_0 before?

    Regards,

    Nikhil

  • Hello Nikhil,

    I modified the app, but I did not modify MCU2_0 and MCU4_0.

    And their CPU % is too high compared to the previous SDK with my same application.

    No, I don't use MCU4_0.

    Best regards,

  • Hi,

    No, I don't use MCU4_0.

    There are changes in the multi-cam app in SDK 8.6 where the 2 VPAC are enabled and the sensors are split among each.

    The 2 VPACs are running on MCU2_0 and MCU4_0 each. 

    You could set the flag "bypass_split_graph = 1" in the config file to use only MCU2_0 similar to SDK 8.5 release.

    Could you please confirm if you are doing this?

    I modified the app

    May I know what are modifications done?

    Regards,

    Nikhil

  • Hello Nikhil,

    Is there an updated document to explain bypass_split_graph = 1?

    I used the previous SDK-based code, so I don't use bypass_split_graph = 1.

    Best regards,

  • Hi,

    I believe this is a GAP in the documentation. I have raised this internally and the documentation shall be updated in the next release.

    Meanwhile, If you do bypass_split_graph = 0,  the application should be same as in the previous SDK release.

    bypass_split_graph flag was implemented inorder to split the cameras equally among two VPACs on J7AHP (One running on MCU2_0 and One running on MCU4_0)

    Making it 0 would make all cameras to run on single VPAC on MCU2_0.

    Regards,

    Nikhil

  • Hello Nikhil,

    I still have a question.

    If the applications in SDK 8.2.2 and SDK 8.6 bypass_split_graph = 0 are identical, I don't understand why those MCU2_0 and MCU4_0 in SDK 8.6 use very high CPU utilization.

    Best regards,

  • Hi,

    Could you please confirm if you are having bypass_split_graph = 0 in your config?

    May I know what are modifications done?

    If there are no nodes assigned to the MCU4_0, then there shouldn't be any load on the same? Are you using MCU4_0 for something else
    Could you please share what are the modifications done to the application?

    Regards,

    Nikhil