0

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.
Prerequisites: Everything was fine yesterday, I didnt install any drivers or updates
A problem: I have bad screen resolution 640x380
Current OS is: Ubuntu 20.04.1 LTS
GPU: geforce gtx 1080
nvidia driver is: 455
cuda is 11.1: but i didnt touch any settings related to it more than a few weeks, so it couldn`t affect it, because i switched on and restarted my PC for many times after installing it and there is not any problems
What I have done to try to fix it:

1)

sudo apt update -y
sudo apt upgrade -y
sudo reboot now

sudo prime-select intel
sudo prime-select nvidia
# i didnt have /lib/modprobe.d/blacklist-nvidia.conf 
# so i wasnt able to delete it
sudo update-initramfs -u
sudo reboot now
  1. I have disabled the security boot in my bios
  2. Tried to install again, but it was written: "I have the latest version"
sudo ubuntu-drivers autoinstall

Any thoughts?

Yuri Glushenkov
  • 101
  • 1
    Did your kernel update to 5.8.0-34-generic yesterday? If so, chances are your NVIDIA driver was installed from NVIDIA themselves and not the graphics-drivers PPA which doesn't install the DKMS driver version that will carry over into the new kernel. Any kernel updates when using the proprietary driver from NVIDIA will have to be reinstalled after the kernel update. – Terrance Jan 07 '21 at 14:45
  • @Terrance Hi As I wrote above I didnt do any updates/installations explicit by hand yesterday. May the os do it automatically? If so how can I check it? My kernel version now is uname -r -> 5.4.0-59-generic Thanks – Yuri Glushenkov Jan 07 '21 at 14:56
  • Try booting to your previous kernel of 5.4.0-58-generic and see if the driver still works there. You may not have the HWE kernel installed which bumps up to 5.8.0-34-generic, but the CUDA installation that installs its own driver does not install the DKMS driver that carries over into the new kernel. I have seen this many times on here. – Terrance Jan 07 '21 at 14:59
  • @Terrance Yes, it works with 5.4.0-58-generic, thanks. What should I do now? – Yuri Glushenkov Jan 07 '21 at 15:19
  • I actually just updated my answer here https://askubuntu.com/a/1288405/231142 on how to install CUDA 11.2 with using the graphics-drivers PPA along with it so that you don't bump into driver issues after kernel updates. Take a look. – Terrance Jan 07 '21 at 15:20
  • @Terrance thanks a lot, I'll do it. I've read it and I'm sure, it'll be enough to download wget https://developer.download.nvidia.com/compute/cuda/11.1.0/local_installers/cuda_11.1.0_455.23.05_linux.run instead of 11.2.0 and change cuda 11.2->11.1 and everything will work. I just have a question why do u use here the 11.1st version instead of 11.2nd (in ur article u use 11.2) export LD_LIBRARY_PATH=/usr/local/cuda-11.1/lib64${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}} and in the section "Install libcudnn8" ubuntu1804 instead of "ubuntu2004"? – Yuri Glushenkov Jan 07 '21 at 15:38
  • Thanks for catching that. Also, the installation for 20.04 is actually no different than the 18.04 other than possibly the NVIDIA driver name. I manually updated that answer for that part by typing in instead of copy and paste and I missed changing that 11.1 to 11.2. If you choose to do the 11.1 installation make sure that you change the parts of the .profile lines there to match the 11.1 installation. – Terrance Jan 07 '21 at 15:41
  • One more thing, I updated that answer one more time to reflect the libcudnn installation that actually still uses the 18.04 and not the 20.04 as it doesn't allow for installation from the 20.04 repo. – Terrance Jan 07 '21 at 15:56

1 Answers1

1

I had exactly the same problem but it's been fixed in today's update.

Just update and it will be solved

amarintj1986
  • 31