Installed TITAN V driver 387.34 but rolled backed automatically to 384.111 after installing CUDA 8.0-GA2

Accelerated Computing NGC GPU Cloud TITAN
#1

I am working on:
Ubuntu 16.04.3 LTS
kernel 4.4.0-109-generic
gcc version 5.4.0 (Ubuntu 5.4.0-6ubuntu1~16.04.5)

I was using GeForece 1050ti and driver 384.111. (I checked by nvidia-smi).
Also CUDA 8.0 GA2 was installed.

After I changed GPU to TITAN V and remove/purge old driver and CUDA toolkit
by “sudo apt-get remove --purge nvidia-" "sudo apt-get remove --purge cuda-”

I installed nvidia driver for TITAN V (387.34).
I checked nvidia-smi says “387.34”

But now I re-installed CUDA 8.0 GA2 took kit,
I checked nvidia-smi syas “384.111”.

I all followed installation guide by
http://docs.nvidia.com/cuda/cuda-installation-guide-linux/index.html#post-installation-actions

DO I NEED TO INSTALL CUDA 9.0 or 9.1 to utilize TITAN V driver 387.34, NOT CUDA 8.0 ?
OR HOW I CAN UTILIZE TITAN V driver 387.34 ?

#2

Sorry that I post this.

I noticed that this is GPU cloud post …

I will find some other topic to post.

#3

You need to use at least CUDA 9.0 for proper Volta support generally. (Note: When you install the toolkit of whatever version, you can tell it not to replace the driver you already have. The exact method for that depends on whether you’re installing from .run or .deb or .rpm, etc.)

That said, are you using the NGC containers? If so, and if you’re not in any other way building CUDA apps from source yourself, you may not actually need to install the CUDA Toolkit as a whole; the parts of CUDA that the NGC containers need are already included in the NGC images, for example. You would only need the driver installed to the bare metal to run those containers.