ubuntu for Jetson TX2

Autonomous Machines Jetson & Embedded Systems Jetson TX2
1

I have questions about ubuntu for Jetson TX2 and tensorflow and my questions are as follow:

  1. According to your jetpack release note, Jetpack provide ubuntu for jetson. At this moment, is it(ubuntu for jetson) as same as general ubuntu? That is, NVIDIA provides a modification version of ubuntu depending on Jetson TX2?

  2. As far as I know, we need to do source build for tensorflow when we start to install tensorflow on TX2. Can you tell me why we have to do source build for tensorflow instead of pip install tensorflow-gpu in installing tensorflow?

I think above questions are basic questions but I really want to know the reason and clarify to know the system. please answer my questions.

2

The “sample rootfs” is purely stock Ubuntu. The “apply_binaries.sh” step overlays drivers with direct hardware acceleration on top of this (it is purely Ubuntu with some drivers to hardware being custom to NVIDIA hardware). Example: The GPU can be used with the NVIDIA driver…without this the Nouveau driver would be required and this does not allow GPU to accelerate video or CUDA.

JetPack is a front end. One thing it does is download the driver package and sample rootfs followed by the “apply_binaries.sh” step. I normally just use the command line and not JetPack unless I am installing extra packages (JetPack can download and install other software to both host and Jetson…command line cannot do this without special efforts).

I do not know an answer for the tensorflow topic, but I would expect the C/C++ API to have different support from a Python API.