I am somewhat new to Ubuntu, therefore I hope that my question is not too silly.

Basically, I need to install NVIDIA_Linux_x86_64-346.72.run on my laptop (acer v5-573g-54208) to use CUDA, cudnn and theano to speed up convolutional neural network computations (using Lasagne).

I am trying to solve this problem for a long time now. I have installed the drivers (NVIDIA_Linux_x86_64-346.72.run), but after logging in to ubuntu my screen turns black and I am back at the log-in window.

If anyone could help me out on this, I would be very, very grateful! Thank you!

2 Answers 2


I haven't had much luck with the the proprietary drivers, so here is a solution using the modified drivers that should work just as well. Make sure to remove any nvidia drivers or cuda installation before following these instructions.

Install the drivers:

sudo add-apt-repository ppa:xorg-edgers/ppa
sudo apt-get update
sudo apt-get install nvidia-XXX nvidia-prime

Restart your computer and make sure the drivers installed correctly. To check, install nvidia x server settings. If your gpu is found, this software will display information about your graphics card.

Next install the cuda toolkit from standard repository:

sudo apt-get install nvdia-cuda-toolkit

Then install the correct cuda library from the repository added earlier:

sudo apt-get install libcuda1-XXX
sudo add-apt-repository -r ppa:xorg-edgers/ppa

XXX in the above code will depend on what graphics card you have. For me (I have an nvidia 970M) it was 346.

Finally, try running deviceQuery (may have to run as sudo) to see if cuda is installed and finding your gpu.


Another possibility is to first install the nvidia driver using apt-get, then download the run file for the cuda version that you need from the nvidia website, and install only cuda and the cuda samples from the run file. For example, for cuda 7.5, this is what I used:

sudo ./cuda_7.5.18_linux.run --silent --toolkit --samples --samplespath=/usr/local/cuda-7.5/samples --override

This allows you to use any nvidia driver that you can install using apt-get with any version of cuda.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.