Platform: Local Server - Ubuntu

I had no problem upgrading 18.04 to 20.04 with these instructions: environment setup with local GPU - Ubuntu 20.04

I’ve a laptop with Intel Core i9, 8 GB RAM (to be upgraded to 16GB after the crisis), Nvidia 1650 with 4GB vRAM. Though there are multiple cloud options, running your notebooks locally feels great, at least for me. So I setup dual boot with Ubuntu 20.04 on my laptop. I ran the first 3 notebooks without any issues (except for the first lesson, the pet dataset takes longer than a while to download). All the information you’re seeing below is already present on the forums and FastAI Github page and course page. I had to do this twice since I messed up my OS setup and I had to search in the forums again to complete this. So I thought of putting up a note for myself and share it with you people. Thanks to @FourMoBro and all my other classmates who pitched in the forums.

Github Links

  1. The Part 1 - 2020 course is running on fastai v2, which isn’t generally available yet. So, we need to install it from the github source (since under active development)
git clone
cd fastai2
conda env create -f environment.yml
source activate fastai2
  1. Once done, cd into the cloned directory and install from the latest code
git clone
cd fastai2
pip install -e ".[dev]"

I made a silly mistake in forgetting to install after creating the environment above. So, when I tried running the course notebooks,it was not able to find the fasti2 library

  1. To use fastai2.medical.imaging you’ll also need to:
conda install pyarrow
pip install pydicom kornia opencv-python scikit-image
  1. fastai2 installation is now done.

  2. For the course/fastbook notebooks to run, there are some extra dependenices that you might want to install

pip install graphviz
pip install azure-cognitiveservices-vision-computervision
pip install azure-cognitiveservices-search-websearch
pip install azure-cognitiveservices-search-imagesearch
pip install "ipywidgets>=7.5.1"
pip install sentencepiece
pip install scikit_learn
  1. Once the installation is done, you can cd into the coursev4 or fastbook and start running the notebooks.

Link to Google Docs version of the same

I’ll try and update the notes with how I installed the Nvidia GPU driver as well.


I have just setup a DL workstation using an RTX 2070 Super and have used Unigene Heaved and Superposition to compare my setup and they seem just about there.

Are there any such standardized deep learning benchmarks I can use? I have been trying to replicate notebooks in Fastbook which I have previously run on Colab Pro and GCP using a Tesla P100, but I don’t think that gives an accurate representation…

Not to the best of my knowledge: Here’s what I did, bench marked individual parts and compared their scores online (after averaging over 5).


Any specific tests you used to benchmark the individual parts? I used the Unigene suite to benchmark the GPU, specifically Heaven and Superposition and compared my results to the leaderboard scores with the similar configs.

pip install graphviz

Thank you very much, worked beautiful (ubuntu 18.04, GeForce 2080Ti, driver version 440.100, CUDA v 10.2)

Hi everyone,
I’m totally new to this topic and i’ve been struggling with the set up part.
I have a GeForce MX150 on an Asus Laptop.
I am unable to install properly the Nvidia driver (tried with 390 and 440 version) and obtain the first picture of this post. I have followed the instructions but get the error :
“NVIDIA-SMI has failed because it couldn’t communicate with the NVIDIA driver. Make sure that the latest NVIDIA driver is installed and running.”
I’ve looked at different forums but most of them seems to reinstall the driver and things work which is not my case.
Does someone encounter the same issue ?
If you need more info on my config or previous installation let me know.