Official TensorFlow for Jetson AGX Xavier

hello @skama these are the commands return

nvidia@jetson-0423618000766:~$ ls -la /usr/lib/aarch64-linux-gnu/libcu*
lrwxrwxrwx 1 root root 16 Nov 22 16:19 /usr/lib/aarch64-linux-gnu/ -> tegra/
-rw-r–r-- 1 root root 51552 Jun 23 01:36 /usr/lib/aarch64-linux-gnu/
lrwxrwxrwx 1 root root 23 Apr 6 2018 /usr/lib/aarch64-linux-gnu/ ->
-rw-r–r-- 1 root root 198584 Apr 6 2018 /usr/lib/aarch64-linux-gnu/
-rw-r–r-- 1 root root 30408 Jun 23 01:36 /usr/lib/aarch64-linux-gnu/
-rw-r–r-- 1 root root 22216 Jun 23 01:36 /usr/lib/aarch64-linux-gnu/
-rw-r–r-- 1 root root 116728 Jun 23 01:36 /usr/lib/aarch64-linux-gnu/
-rw-r–r-- 1 root root 526152 Jun 23 01:36 /usr/lib/aarch64-linux-gnu/
lrwxrwxrwx 1 root root 19 Oct 29 20:10 /usr/lib/aarch64-linux-gnu/ ->
lrwxrwxrwx 1 root root 23 Oct 29 20:10 /usr/lib/aarch64-linux-gnu/ ->
-rw-r–r-- 1 root root 436464 Oct 29 20:10 /usr/lib/aarch64-linux-gnu/

nvidia@jetson-0423618000766:~$ update-alternatives --display libcudnn
update-alternatives: error: no alternatives for libcudnn

Hi, artemisl

Please install CUDA/cuDNN package directly from the JetPack.
The package you downloaded is for desktop environment(x86), doesn’t support Jetson platform(aarch64).

There is a known issue that JetPack installer may close after flashing without installing packages in several case.
Please re-check if you have well-installed the package or please install the package again.


Hi, AsataLLL

There are some error, when i flashing Xavier with JetPack installer.
The JetPack installer didn’t install CUDA/cuDNN

@artemisl, It looks like you are having trouble installing jetpack rather than the tensorflow. Is it persistent that if you try to flash a couple of times all of them fail to install cuda/cudnn?


It looks like CUDA cannot be installed on your device.
Could you help to check the network functionality first?


Hi artemisl,
Your problem seems like some network setup problem. When you flashing Xavier with JetPack installer, there is a page that let you select the network card on the host with internet access, that is a serious step to make sure your Xavier has the right internet access. Hope it helps.

Thank you for everyone help me to solve the problem.

The problem is JetPack 4.1 DP can’t fully install all tensorflow all dependence. It needs to use apt-get install on Xavier’s device to install some library or package.

Does this work if you install miniconda on Jetson? And is the problem artemisl brought up a non-issue at this point?

Sorry that we don’t have an experience of it.
You can give it a try and file a topic if any problem.



I try to follow post #1 to install tensorflow after flashing Xavier.

It got stuck at the message below:

Collecting h5py (from keras-applications>=1.0.6->tensorflow-gpu==1.12.0+nv19.1)
Downloading (287kB)
100% |████████████████████████████████| 296kB 5.4MB/s

Nothing happens after 100%… Can anyone help?




The installation takes time.
Please wait and it will go next steps after several minutes.


Could you please provide a build of the library and associated headers so we can use TensorFlow in C++ as well?


Sorry that we don’t provide TensorFlow C++ libraries due to the limited resource.
You can check the community for help.


I just install Jetpack4.1.1 on my nvidia xavier and then wanna install the tensorflow on the board.Thus I follow the official docunment:

Unfortunately comfront some errors.
Firstly,I install Nvidia SDK Manager:
$ sudo apt install ./sdkmanager_0.9.11-3405_amd64.deb
Some packages could not be installed. This may mean that you have
requested an impossible situation or if you are using the unstable
distribution that some required packages have not yet been created
or been moved out of Incoming.
The following information may help to resolve the situation:

The following packages have unmet dependencies:
sdkmanager:amd64 : Depends: libgconf-2-4:amd64 but it is not installable
Depends: libcanberra-gtk-module:amd64 but it is not installable
E: Unable to correct problems, you have held broken packages.

$ sudo apt install libgconf-2-4
libgconf-2-4 is already the newest version (3.2.6-4ubuntu1).
The following packages were automatically installed and are no longer required:
libdbusmenu-gtk4 libdbusmenu-qt5-2 libgsettings-qt1 liblockfile-bin
liblockfile1 libqt5sql5 libqt5sql5-sqlite lockfile-progs x11proto-dri2-dev
Use ‘sudo apt autoremove’ to remove them.

The dependencies are exist,but I still couldn’t install SDK. Why?

hi all,

I am very sorry for late report the problem status.
I solve the Problem by check all network connect correct when reflash JetPack 4.1 DP
The JetPack installer will install CUDA/cuDNN

After that, the tensorflow can install correctly


Have u run into the problem below when u installed the tensorflow on Xavier:

$ pip3 install --extra-index-url tensorflow-gpu==1.13.0rc0+nv19.2 --user

Traceback (most recent call last):
File “/usr/bin/pip3”, line 9, in
from pip import main
ImportError: cannot import name ‘main’

Due to this,I failed to install the tensorflow correctly.

Appreciate to get your reply.Thanks.


Please try the change in this comment to fix the pip issue:


Hi AastaLLL,

One more trouble,

~$ pip3 install --extra-index-url tensorflow-gpu==1.13.0rc0+nv19.2 --user
bash: pip3: command not found

~$ sudo apt autoremove libdbusmenu-gtk4 libdbusmenu-qt5-2 libgsettings-qt1 liblockfile-bin
Reading package lists… Done
Building dependency tree
Reading state information… Done
The following packages will be REMOVED:
libdbusmenu-gtk4 libdbusmenu-qt5-2 libgsettings-qt1 liblockfile-bin
liblockfile1 libqt5sql5 libqt5sql5-sqlite lockfile-progs x11proto-dri2-dev


pip3 can be installed by this command:

sudo apt-get install python3-pip