This is on CentOS 7. I had the 384.111 driver already installed.
When I installed the cuda toolkit (via rpm), it did not ask if I wanted the driver installed.
It appears to have done so anyway - and improperly. nvidia-smi is gone, and running it from the extracted driver folder gives a “driver/library version mismatch” error.
So, did it overwrite the driver? (I assume yes).
Is there a way to make the rpm install not install the driver, or is that only achievable via the “run” install?
if you read the linux install guide, you can learn how to install the cuda-toolkit without the driver, even using a package manager install method
I’ve read the relevant sections. I must be missing it … there does not appear to be such an option for the rpm install.
read all of section 3.7 carefully.
in particular section 3.7.3
in particular table 4
in particular this:
cuda-toolkit-9-1 Installs all CUDA Toolkit packages required to develop CUDA applications. Does not include the driver.
(when I used a simple search function on the linux install guide to find cuda-toolkit, it takes me to that line)
yum install cuda-toolkit
should install the latest that is available in the repo
yum install cuda-toolkit-X.Y
should install the CUDA toolkit X.Y version (assuming version X.Y is available in the repo)