Cuda driver version and runtime version not matched

I installed cuda toolkit 11.4.1 on windows. But somehow the runtime library is still pointing to the old version v10.2. how can I change to match the driver library? Thanks,

image

Follow the instructions in the linux install guide, carefully, especially step 9:

https://docs.nvidia.com/cuda/cuda-installation-guide-linux/index.html#post-installation-actions

Also, the CUDA sample codes will display the version of CUDA toolkit (Runtime) that they were compiled with. If you have not recompiled your sample codes, and run the newly compiled versions installed by the 11.4 installer, you won’t get a runtime report of 11.4 here.

Thanks for the reply,

I have tried to recompile in Microsoft visual studio. But it is calling nvcc (10.2) instead of nvcc (11.4). My feeling is that I missed some environmental parameters.

CUDA_PATH does point to the newly installed toolkit (11.4).

In visual studio you need to make sure you are building with the CUDA 11.4 toolchain. You can research how to do that if you wish. I generally don’t try to use multiple toolchains in VS, so one possible solution is to remove and reinstall Visual Studio, then reinstall the CUDA 11.4 toolkit installer.

It’s also remotely possible (I suppose) that you are using an unsupported version of VS for CUDA 11.4, so the 11.4 installer may have not installed the toolchain into VS.

I didn’t realize you were on windows initially, so the correct installation guide is here:

https://docs.nvidia.com/cuda/cuda-installation-guide-microsoft-windows/index.html

This is really an installation issue, not a programming issue, and there is a separate forum for that here:

Finally got it working. thanks,