Errors while installing TensorFlow on Jetson Nano

Hello all,

I’ve followed the instructions on: https://docs.nvidia.com/deeplearning/frameworks/install-tf-jetson-platform/index.html#install

to install Tensorflow with:
sudo pip3 install --extra-index-url https://developer.download.nvidia.com/compute/redist/jp/v43 tensorflow-gpu==1.15.0+nv20.1

There were no errors on installation . But when I try
import tensorflow
from python3 cmd line, I get the following message:
2020-08-17 22:01:48.200850: W tensorflow/stream_executor/platform/default/dso_loader.cc:55] Could not load dynamic library 'libcudart.so.10.0'; dlerror: libcudart.so.10.0: cannot open shared object file: No such file or directory; LD_LIBRARY_PATH: /usr/local/cuda-10.2/lib64:
2020-08-17 22:01:48.200936: I tensorflow/stream_executor/cuda/cudart_stub.cc:29] Ignore above cudart dlerror if you do not have a GPU set up on your machine.
2020-08-17 22:01:48.201172: W tensorflow/stream_executor/platform/default/dso_loader.cc:55] Could not load dynamic library 'libcudart.so.10.0'; dlerror: libcudart.so.10.0: cannot open shared object file: No such file or directory; LD_LIBRARY_PATH: /usr/local/cuda-10.2/lib64:
2020-08-17 22:01:48.201218: I tensorflow/stream_executor/cuda/cudart_stub.cc:29] Ignore above cudart dlerror if you do not have a GPU set up on your machine.
Segmentation fault (core dumped)

Please let me know how to deal with this issue

Hi,

Do you use JetPack4.3?
The command you shared is for JetPack4.3 environment so the TensorFlow is looking for the CUDA 10.0.

If you are using our latest JetPack4.4, which is CUDA 10.2, please use following command instead:

TF-2.2.0

$ sudo pip3 install --pre --extra-index-url https://developer.download.nvidia.com/compute/redist/jp/v44 tensorflow==2.2.0+nv20.7

TF-1.15.3

$ sudo pip3 install --pre --extra-index-url https://developer.download.nvidia.com/compute/redist/jp/v44 tensorflow=1.15.3+nv20.7

Thanks.

cat /etc/nv_tegra_release
gives:
# R32 (release), REVISION: 4.3, GCID: 21589087, BOARD: t210ref, EABI: aarch64, DATE: Fri Jun 26 04:38:25 UTC 2020
So, is the JP version 4.3?

But nvcc --version
gives
Cuda compilation tools, release 10.2, V10.2.89

P.S:

  1. I’m trying to install TF on my Nano. And, the above commands are run on nano itself.
  2. But, when I connected my Nano to my host machine & opened SDKManager, it is showing that JP4.4 is installed(attaching screenshot)

I’m confused. What version of JP is installed on my Nano?

Thanks a lot for your time!

Update:

I used
$ sudo pip3 install --pre --extra-index-url https://developer.download.nvidia.com/compute/redist/jp/v44 tensorflow==1.15.3+nv20.7
& was able to install TF.

But, why did
cat /etc/nv_tegra_release
show JP verison as 4.3?

Hi,

Based on your image, the JetPack version you used is v4.4.

The information shown in nv_tegra_release is the OS version rather than JetPack version.
(sorry that this might bring some confusing)

#R32 (release), REVISION: 4.3, GCID: 21589087, BOARD: t186ref, EABI: aarch64, DATE: Fri Jun 26 04:34:27 UTC 2020

The information indicates it is r32.4.3, the OS image includes in JetPack4.4.
https://docs.nvidia.com/jetson/jetpack/release-notes/index.html#os

Thanks.

Alright, thanks!

Also, is there a way to check JetPack version, other than the way i used above (connecting the target device to host device & launching SDKManager & letting it detect automatically)?

Hi,

You can check software version directly.
For example, r32.4.3 indicates that you are using JetPack4.4 product release (GA).

Thanks.

1 Like