• State TI Thinks Resolved
  • Locked Locked
  • Replies 3 replies
  • Answers 1 answer
  • Subscribers 99 subscribers
  • Views 142 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.

TDA4VEN-Q1: Does the model inference memory usage of the TDA4 entry-level J722S (SDK11) TIDL node on C7x_1 decrease compared to TDA4VM (SDK10)?

Q xn
Prodigy 180 points
Part Number: TDA4VEN-Q1
Other Parts Discussed in Thread: TDA4VM

Tool/software:

Hi：TI

 Recently, I tested the TDA4 entry-level SDK11 (J722S) and found a significant reduction in memory usage for TIDL model inference on the C7x_1 core compared to TDA4VM SDK10. Was this due to optimizations? Below is my data comparison between the two platforms.TDA4VM c7x_1 cpu 21M .TDA4 entry c7x_1 cpu 4.48M

 

  • 0
    TI__Expert 4820 points

    Hi Q,

    Yes the 11.0 SDK release has some TIDL optimization and C7x/MMA code updates as well that contribute to the change. Did you see a difference between the SDK 10 and the SDK 11 for the J722S? or between the SDK 10 and SDK 11 for J7ES(TDA4VM)? 

    Warm regards,

    Christina

  • 0 in reply to Christina Kuruvilla
    Prodigy 180 points

    Hi :

    On SDK10, we only conducted performance tests for TIDL and did not delve deeply into the specific code.

  • 0 in reply to Q xn
    TI__Expert 4820 points

    Hi Q, 

    You can't really compare two different versions of the SDKs with two different devices, since the J722S and the J7E2 (aka TDA4VM) have different specs. I recommend checking with TIDL 11.0 for the J7E2/TDA4VM and seeing if there is a change there.

    Also, you don't need to dive into the code. The 11.0 SDK has updates regarding the C7x/MMA that can impact TIDL.  

    Warm regards,

    Christina