CUDA 9.0 driver vs GeForce Driver

I have successfully setup CUDA 9.0 and cudnn 7.0.5 on my Windows 10 machine for Tensorflow and Keras. TF currently only supports CUDA 9.0 so if I update the Geforce driver from Geforce Experience, the latest being 397.31, will it break the CUDA 9.0 installation? Are the Geforce drivers tied to specific versions of CUDA?

Secondly, which components do I need to install in Visual Studio 2017 to run CUDA successfully? VS2017 has tons of components and installing all of them seems to be overkill.


Installing a newer GPU driver in a functional CUDA 9 setup should not break anything.

The primary component you need in VS 2017 is Visual C++