Did anyone get into some issues with the installation of centos 7 CUDA 10.1 package (tried both with rpm and runfile) with the recent NVIDIA drivers 430.xx (I have tried with 430.26)? It seems to be some incompatibilities: during the installation some components of older drivers (418.67) are required, which obviously can not be installed with the new drivers. I also noticed that rtx 2080ti card is not supported by 418.67 drivers.
I was suggested to install just cuda-toolkit rather than the whole package and the installation went fine, but then it seemed to cause troubles during gpu based applications.
Dear all,
in the meantime I had to re-install the OS; then I tried to re-install NVIDIA drivers suggested for my cards (RTX 2080 ti) by the web site (430.64) and everything (nvidia-smi) seemed OK; I then installed cuda package (10.1 cuda-repo-rhel7-10-1-local-10.1.168-418.67-1.0-1.x86_64.rpm) and the installation went fine, but after that, when trying to fulfil devicequery and bandwidthtest I got “cudagetdevicecount returned 999” and “cudagetdeviceproperties returned 999”.
Also running nvidia-smi after the installation of cuda gave “nvidia-smi failed to initialize nvml driver/library version mismatch”. I then installed NVIDIA drivers 418.67 and everything seems to work properly now. Maybe this is just the way it must be, I had figured out I could have installed the latest NVIDIA drivers and the latest CUDA package.
Your opinions/hints will be appreciated,
thanks
Stefano