Hi,
Could you check if TensorRT works when putting the model into DLA first?
Since not all the operations in TensorRT are supported by the DLA, you might need to enable allowGPUFallback to enable the pipeline for certain model.
If above works well on your environment, could you share a photo of your Nsight system with us?
Suppose you should be able to see the DLA status at attached picture.
Please also noticed that Nsight System only output the DLA status (idle or busy) rather than utilization.
Thanks.
