One day booting my PC I noticed that only one out of two monitors displayed the signal/data from the PC. Switching cable, ports and rebooting I realized that every of the three display ports of my trx 2080 ti were working fine but not at the same time. At maximum one port/display shows the PC’s output data. Further when calling nvidia-smi I had the following result: “NVIDIA-SMI has failed because it couldn’t communicate with the NVIDIA driver. Make sure that the latest NVIDIA driver is installed and running.”
Going to “settings”->about I found out that the Graphics are managed by “llvmpipe (LLVM 7.0, 256 bits)”. Seems to be the fallback variant of the Mesa 3D graphics library when no GPU driver is available. Ubuntu 18.04.2 is currently installed on my machine.
Reinstalling the NVIDIA driver several times and using different methods (e.g. by executing run_file {NVIDIA Driver, CUDA 10.0 including driver installation}, using the graphics driver ppa from Ubuntu, .deb packages for CUDA 10.0 and CUDA 10.1 including driver installation) did not solve the problem.
After removing the old driver and re-installing the latest versions of the driver (418 and 410) I will get always the same message when executing nvidia-smi:
“NVIDIA-SMI has failed because it couldn’t communicate with the NVIDIA driver. Make sure that the latest NVIDIA driver is installed and running.” Further, only one monitor works while the other does not a signal.
I also tried another GTX graphic card (different model) and got the same issue.
The strange thing is, that I did not change anything on my system when problem has occurred the first time, besides some regular Ubuntu updates. Could those have been affected the NVIDIA driver?
Please run
grep nvidia /etc/modprobe.d/* /lib/modprobe.d/*
to check if a file exists containing
blacklist nvidia
if so, please delete it and run
sudo update-initramfs -u
then reboot.
Running grep nvidia /etc/modprobe.d/* /lib/modprobe.d/* I get the following result:
/etc/modprobe.d/blacklist-framebuffer.conf:blacklist nvidiafb
/etc/modprobe.d/disable-nouveau.conf.backup:blacklist nvidiafb
/etc/modprobe.d/nvidia-installer-disable-nouveau.conf:# generated by nvidia-installer
Those files are correct, please leave them in place.
The problem is that the nvidia driver is correctly installed but the kernel modules don’t load. Most common reason are either a blacklist file or secure boot being enabled but both don’t apply to your system.
Please post the output of
modinfo nvidia
and try to load them manually
sudo modprobe nvidia
Please post any error that is displayed.
In meantime I have found out that my cc and c++ refer to clang-8 and clang+±8. I used the update alternatives in order to clang the default compiler. I thought that maybe that has caused the issue. I changed the update alternatives of cc and c++ to refer to gcc-7 and g+±7. Then I removed Cuda and Nvidia with purge and reinstalled Cuda 10.0 based on the network .deb file. Calling nvidia-smi after that lead to the correct result. I think the problem has been originally caused by cc and c++ referring to Clang.
Might very well be, though it’s still odd why the dkms makefile didn’t catch the compiler mismatch.