MXNet Forum

Mxnet-tensorrt integration on the jetson tx2


#1

Has anyone used the tensorrt integration on the jetson.

I’m getting build errors relating to not finding onnx. Onnx has been installed and I tried mapping it in a few different ways.

Is the integration affected by the jetson not supporting the tensorrt python api?


#2

Hey @nchafni. It’s quite difficult to build TRT manually on the TX2 at the moment. What I would recommend is to follow the installation steps in the CI very closely. If you first build ONNX, then ONNX-TensorRT it should be the easiest way to verify that you have the correct dependencies.

You would need to substitute CUDA_ARCH="-gencode arch=compute_62,code=compute_62" instead of CUDA_ARCH="-gencode arch=compute_70,code=compute_70"