NVIDIA CUDA Toolkit and cuDNN deployment in production

Hi,

I am working on a deployment process for a deep learning solution in the Windows environment.
Because my deep learning solution is based on NVIDIA GPU hardware, so the utilization of CUDA Toolkit (V9.0) and cuDNN (7.1) are necessary and I want to deploy them through my own deployment process for Windows platform only.

I found these two links below about CUDA and cuDNN distribution respectively:

My problem is that the distribution or deployment process in these two links is not clear enough, so I am not sure if I have the right to:

  • take DLLs from CUDA Toolkit instead of the whole installation package, so I can avoid installing this software with the standard way and save disk space because I do not need all DLLs in this toolkit for sure.
  • in the same way for the cuDNN package, if I can only take some files in it and install it automatically within my installation process.

Do you have an idea how to be sure that what I want to achieve is legal?

Thank you very much.
Best regards,
Gaël