Two Days to a Demo Tutorial (Solved)

Hi,
I have some difficulties to play with this tutorial.
My laptop is on Ubuntu 18.4 Bionic. Is it good for this tutorial?
Should I use python2.7 instead python3 for Digits?
Lastly, this ‘sudo apt-get install caffe-cuda’ doesn’t work for me.

fyi : I have installed cuda driver, cuda, Jetpack without problems.

Thanks in advance.

Hi ndnparis, the training host PC for this tutorial is verified with Ubuntu 16.04.

Also, which GPU does your laptop have?

Hi Dusty,
Thanks for your reply.
My laptop has GTX1600-6Go and it works well with CUDA Driver version 410.66.
Should I downgrade my laptop to Ubuntu 16.04 for this tutorial?

You should be able to compile Caffe from source if you prefer to not downgrade, by following this part of the tutorial.

Note that the Building DIGITS instructions are for Ubuntu 14.04 / 16.04, so I am not sure how it will work with 18.04.

Hi Dusty, I am downgrading my laptop to 16.04 because I have difficulties to install caffe and digits.
Thanks for your help.

Hi Dusty,
Under CUDA 10 and cuDNN 7.3, CAFFE works great with DIGITS.
As you know, currently DIGITS targets tensorflow v1.2.1, CUDA 9 and cuDNN 5.1. So could you tell me if there is a plan/roadmap for CUDA 10 and cuDNN 7.3? Because I would like to make a little try with Tensorflow, too.
Btw, could you share your experiences/recommendations about CAFFE and Tensorflow? In the case of the Jetson Xavier, which one will be better?
Thanks in advance.

This particular tutorial uses Caffe for ease-of-use with DIGITS, however here are some other repos that show importing TensorFlow model into TensorRT on Jetson:

[url]https://github.com/NVIDIA-AI-IOT/tf_trt_models[/url]
[url]https://github.com/NVIDIA-AI-IOT/tf_to_trt_image_classification[/url]

Note that on the Jetson side for Jetson Xavier, they are all using CUDA 10 / cuDNN 7.3 with JetPack 4.x.
The CUDA9 version is for the host training with DIGITS, although it may be possible to upgrade that.

Ok, me to I’m trying to use on Xavier the tf_trt_models and tf_to_trt_image classification.

But, may not directily related to Xavier new Ubuntu 18.04 version and DevKit component version (CUDA 10, cuDNN 7.3), I’m not able to use them, in particular:

[b]
$ make
==> missing libcublas.so.9.0

at tf_to_trt_image_classification
[/b]

Any help, please?

@Dusty : Thanks a lot for these links.

Hi,

Xavier is using CUDA 10.0.
Please use the TensorFlow package on this link:
https://devtalk.nvidia.com/default/topic/1042125/official-tensorflow-for-jetson-agx-xavier/

Thanks.

@AastaLL :thanks.

A video walkthrough of natively installing NVIDIA DIGITS on Ubuntu 18.04 LTS is available here:

It goes over compiling NVCaffe, all the requirements needed etc.

Hope this helps!

-Cuda Education