Quadro M2000M outperform Xavier in TF

Autonomous Machines Jetson & Embedded Systems Jetson AGX Xavier
1

Hello,

We have tested an example “TensorFlow For Poets” ( [url]https://codelabs.developers.google.com/codelabs/tensorflow-for-poets/#0[/url] ) with customs images and the notebook with NVIDIA M2000M and i7 6820HQ outperformed Xaver. Time on notebook was about 2s on Xavier it was 5s. Considering article “EmBench: antifying Performance Variations of Deep Neural Networks across Modern Commodity Devices” we have expected that Xavier will be way faster than old notebook. Jetson is set to MAXN 0. Does anyone know what the issue might be?

2

It’s likely the model uses some layer that’s not well supported by the TensorFlow support for the Xavier.
Can you port it to TRT?

3

Thank you for the reply. I’ll try to find out how to port it.

4

I know I’m asking to much, but is there any recomanded tutorial on how to convert frozen model to TensorRT? All of my attempts to do that failed today.