Hi
I have only a ssh access to a Dell R740 Poweredge Server with a Nvidia Tesla P100 GPU and Ubuntu 18.04.
I wanted to know what was the recommended process to install mostly the drivers and toolkits necessary for Tensorflow without the risk of a login loop or a black screen. (It already happened once with the one from the Ubuntu repo, ssh was not responding, the IT department reinstalled everything so no nvidia-bug-report available. And since I do not have direct access it is not an easy problem to solve without asking IT to do the whole process again).
I have seen some people having difficulties like in here : [url]https://devtalk.nvidia.com/default/topic/1046157/linux/ubuntu-16-04-gui-login-loop-after-installing-nvidia-driver/1[/url]
Thank you to anyone kind enough and taking the time to help me,
It might be questionable why there’s a gui running on a probably headless server and why installing a graphics driver keeps sshd from starting but to have a least intrusive install,
- download the 418.56 .run installer from here: [url]https://http.download.nvidia.com/XFree86/Linux-x86_64/[/url]
- run it with options --dkms --no-opengl-files --no-x-check -Z
- reboot
- download cuda 10.0 .deb
- do first three steps of install instructions
-
don’t install cuda
- instead, run sudo apt install cuda-toolkit-10-0
Thank you very much.
When you say “do first three steps of install instructions” where can I find this instruction, do you mean the first three items of your instructions or another entirely ?
Sorry, forgot three words:
- do first three steps of install instructions on download page
Addendum: according to other user’s problems, Tensorflow seems to be linked against cuda 10.0 so it doesn’t work with cuda 10.1. Download the 10.0 .deb (network) from archives:
[url]CUDA Toolkit 10.0 Archive | NVIDIA Developer
Thank you very much for your time, I will try that as soon as I can.
Thank you very much it worked for me but I had to remove the nouveau driver and purge them first