How to use GPU option in nvvideoconvert plugin for AGX Xavier platform


I’m designing the gstreamer architecture with Deepstream SDK 5.0.
Unfortunately, it does not reach the expected framerate.

The most suspicious part is the vision accelerator, which seems to be used by nvvideoconvert. (convert UYVY to NV12 format)
Using Nsight system profiler, I can get processing time in vision accelerator. It took about 26 ms.

I tried to use ‘compute-hw’ option in nvvideoconvert by 1(GPU option). but the pipline gives Internal data stream error.

Is there any way to improve it?

There is hardware converter engine(named VIC) in Xavier so the execution is not on GPU. For optimal performance , please follow the steps in this post and try again:
Nvvideoconvert issue, nvvideoconvert in DS4 is better than Ds5? - #3 by DaneLLL

Thank you for your reply.

Even though I set the freq. of VIC to max, the performance doesn’t improved.
Perhaps I’m unintentionally using the VIC as maximum freq.

Does the ‘MaxN’ mode affect VIC freq. ?

Thank you.

If your release is Jetpack 4.6, the VIC frequency is shown in tegrastats. Please execute sudo tegrastats and check if it keeps at max frequency always.

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.