I have a Ubuntu 18.04 machine that I am trying to build CUDA on. When I install the driver for the RTX 2070 Super from ppa everything is ok. Things only start going sideways after I install CUDA and reboot the machine. I’m using driver 430.26 and installing CUDA 10.0. Upon reboot, the machine shows that is failed to start the NVIDIA Persistence Daemon and retries several times or then says starting bpfilter.
I am in the process of doing a clean install of everything including Ubuntu to rule out that I have some misconfiguration from earlier when provisioning this machine. To recover from the boot I have to boot in recovery mode to remove and purge cuda via apt-get.
Any help would be greatly appreciated! Let me know if there is any information missing to help troubleshoot the issue. Thanks
If you’re happy with the 430.26 driver install from ppa, then install CUDA without installing the driver, i.e. leave your ppa driver intact. Both the runfile installer and package manager installer methods have options to install CUDA without installing the bundled driver.
“happy with the driver install” means that you installed the driver, did not install CUDA, and you are happy with the functioning of your display, with ordinary workloads and reboots. If the ppa driver meets your needs in that way, then I would attempt to install CUDA on top of that driver, without installing the driver that comes bundled with the CUDA installer.
On the package manager method, the thing to do would be to install cuda-toolkit-10-0 instead of cuda
On the runfile installer, there is an option to install (or not) the driver. Just deselect that option.
This is further covered in the CUDA linux install guide and on many questions on this forum.