How run the driver installer using the extracted files?

Nvidia binary run package not has option to disable install CUDA (encode and decode files) , but is possible extract all files.
How start the installer using command line inside of the extracted folder ?
I want remove all cuda files and after install a previous version.

Thanks for reply.

https://docs.nvidia.com/cuda/archive/11.3.1/cuda-installation-guide-linux/index.html#runfile-advanced

As you insist on the .run file…
Best to use the regular .run driver installer only (not the cuda .run file installer) first. But if you want for whatever reason to use the cuda .run file installer, use the --driver parameter.
Then use the cuda .run file installer with the --toolkit parameter.
That way you can mix driver and cuda versions using the .run file.

“As you insist on the .run file…”
because is for me a better solution than distro driver … all fault from Nvidia in adding wrong cuda version in 470 instead only in 480.

The link above is about cuda installer.
I not wait use cuda run package because I has the cuda 11.3 files.
I want manually remove cuda files in extracted folder and after install distro cuda deb files.

I want only a command for nvidia driver installer in extracted folder.

Complete nonsense, the distro package is better suited for the system. Nvidia in their cuda .run file ALWAYS release the newest driver with the newest driver. Just because your stoneage GPU is not supported anymore does not make it an error on their side!

https://download.nvidia.com/XFree86/Linux-x86_64/470.74/README/installdriver.html
How about you start reading yourself for once!

That sentence makes no sense!
But you wanted cuda 11.3, so use that .run file to get it, damn it!

Whatever man, you insist on .run file first, now you insist on .deb.
I give up on trying to help you, you are so beyond any reasoning!
Good luck

“Complete nonsense, the distro package is better suited for the system. Nvidia in their cuda .run file ALWAYS release the newest driver with the newest driver. Just because your stoneage GPU is not supported anymore does not make it an error on their side!” "“As you insist on the .run file…”
Well the because is using nvidia binary run package is simple to install in offline.
machine.

“Complete nonsense, the distro package is better suited for the system”
true , but for online.

“Just because your stoneage GPU is not supported anymore does not make it an error on their side!”
Oh yes … was long the support … but need say again …
Nvidia had added a cuda version breaking Kepler being only a one next version not having support to Kepler ?
Nvidia will support Kepler in future 470 updates , but each update NOT will allow use CUDA or NVENC.

Thanks for each reply … had good information.
Have a nice day.

You can just run ./nvidia-installer directly inside the extracted archive directory.

If you want to mess with the list of files it installs, you can edit the .manifest file in the same directory before running the installer. If you really want to get into the nitty-gritty details of how things are installed the nvidia-installer source code is on GitHub.

If you run in to problems and post here, please let us know that you altered the driver package or installer so we don’t assume you’re using the stock installer package.

Thanks very much for your very good replies.
I not wait post any issues when trying run a custom nvidia installer.
Unhappily I need figure a “fix” for driver 470 because is installed cuda 11.4 breaking Kepler gpus when using cuda or nvenc for video being a fault from Nvidia.
I not want try edit the nvidia installer , but not has another option in moment. I need figure which files are related with cuda and nvenc (compute, enconde and decode ?) and remove from installer and figure how install use cuda files from driver 465.31.
That issue is creating problems for users.
Sincerely I try not believe Nvidia will create the last driver for Kepler with a wrong cuda not compatible with kepler.
Nothing against drop support for Kepler, but I not want see that fault less yet from Nvidia.

@ aplattner

I not want to try replace cuda files in driver installation folder, but I not see other solution if trying use current nvidia binary driver.
Thus I had edited the manifest file changing file naming for all files below
nvidia-cuda-mps-control
nvidia-cuda-mps-server
libOpenCL.so.1.0.0
libcuda.so.465.31
libnvcuvid.so.465.31
libnvidia-compiler.so.465.31
libnvidia-encode.so.465.31
libnvidia-ml.so.465.31
libnvidia-opencl.so.465.31
libnvidia-opticalflow.so.465.31
libnvidia-ptxjitcompiler.so.465.31
libvdpau_nvidia.so.465.31
nvidia-cuda-mps-control.1.gz
nvidia.icd

The install works , but cuda and nvenc not work.

Trying not editing manifest and renaming all files above to naming 470 not works.

I need say thanks for Nvidia for long support for Kepler GPUs , but creating the last driver for Kepler not being possible use cuda or nvenc was wrong.
Nvidia need fix it.

Kepler users not use driver 470. use driver 465 instead.