Cuda driver version is insufficient for cuda runtime version

Hi, everyone

I installed cuda 9.0 and NVIDIA driver on RHEL 7 server with 4 Tesla P100 gpus
but I got this error message when I checked whether tensorflow used gpu or not

After installing driver and running ‘nvidia-smi’,
it showed 'driver version: 384.183’

Running ‘nvcc --version’ showed ‘Cuda compilation tools, release 9.0, v9.0.176

To install cuda driver version sufficiently, which parts should be checked first?

Sorry if this isnt’ the right place to post this

Thanks for the help

What is the error message?!?!
The versions you listed are enough to run a P100 on RHEL 7.

Rather than try and diagnose what is going on, its usually easier just to install the latest driver for your GPU.

Very often TF can bring its own version of the CUDA toolkit, apart from the one you installed, such as if you install TF via conda. In this case, you will need a more recent driver.