Run a Neural Network in DLA only (without GPU Fallback)


I am using Nvidia Jetson Xavier and Jetson Orin. My aim is to run a neural network completely on DLA without GPU Fallback

I read the requirements in DLA Tutorial and understood that, DLA can handle fully connected layers.

I followed following steps to run the network on DLA only.

  1. Created a neural network in Google colab to predict the output value equation y = 2x+1 and verified its output.
  2. Exported the network to ONNX
  3. used trtexec to export ONNX to engine file.

Relevant Files

I used google colab to build the network, and link for the notebook is below

Steps To Reproduce

I used below command to export the ONNX to engine file

trtexec --onnx=model_2x_plux_1.onnx --saveEngine=try_model_2x_plux_1.engine --exportProfile=model_gn.json --int8 --useDLACore=0  --useSpinWait --separateProfileRun > model_gn.log

Then I am getting below error saying "fc1.weight is not supported on DLA but GPU fallback is not enabled

Can someone please guide me?