Problem getting TensorFlow to work

Autonomous Machines Jetson & Embedded Systems Jetson AGX Xavier
1

Hi,

I just received my Xavier Dev Kit and I am trying to get Tensorflow to work. For installation I followed the instructions given here:
https://docs.nvidia.com/deeplearning/dgx/install-tf-xavier/index.html
The setup finishes successfully but when I try to import tensorflow in python I get the error:

ImportError: libcublas.so.10.0: cannot open shared object file: No such file or directory

I looked through the filesystem and could’t find the usual installation directory for cuda (i.e. /usr/lib/cuda) or the specified file at all. So I reflashed the device and with JP 4.1 and made sure the cuda option was set to install. However the problem remains. Am I missing something? Any advice on how to proceed would be helpful.

2

Check “/usr/local/cuda/” (“/usr/local/cuda-10/”).

3

There is no such directory so I guess CUDA isn’t installed. I figured that CUDA would be preinstalled. But I even flashed the device and selected CUDA sdk explicitly.

4

Hi,

Do you get any error when installing CUDA with JetPack?
If yes, please share it with us.

Besides, could you just try to reflash your device again?
This issue should be fixed if CUDA can be well-installed after reflashing.

Thanks and please let us know the results.

5

I got it working eventually. The problem was that the installation of CUDA failed because my board didn’t have an internet connection at that point. I missed that some dependencies would have to be downloaded. Connecting both host and board to a router solved the problem. TF is working like a charm now.

Thanks for the help.