Just installed nvidia legacy driver v.470 for an old kepler GPU on linux ,seeking proper way to setup cuda

i have just installed the v.470 nvidia legacy driver on linux for GTX 770 which is kepler gpu.
didn’t setup any cuda yet
but ‘nvidai-smi’ out put like this
NVIDIA-SMI 470.223.02 Driver Version: 470.223.02 CUDA Version: 11.4
what mean by saying CUDA Version: 11.4 ? but i didn’t installed any cuda yet.
then , what is the latest cuda version available i could use for my current gpu and driver ‘GTX 770 with driver version 470’ assume it’s version 11.4 maximum?
other question can i setup cuda 10.1 or 10.2 for instance it will just work fine ?
other question , if i’m using conda environment for managing python packages environment , aiming to compile torch from source first to make it work with the GTX 770 ‘which is has cuda compute capability level3’ , should i install cuda through the conda environment only ? or aside with installing cuda on system wide ‘the proper version that i asked what is it at first’ ?
thanks in advance , help me end this struggle to solve my problem .

The Cuda version shown in nvidia-smi indicates the last version of Cuda the driver will support.

Cuda 10.2 is the last version to support CC 3.0. There’s a useful table at the beginning of the “GPUs supported” section here.

I know nothing about torch or conda, so can’t offer anything more.