2

I'm sorry if I'm asking a silly question. I'm new to CUDA. Installed CUDA 10.1 using Runfile method and did the following export according to Nvidia instructions:

export LD_LIBRARY_PATH=/usr/local/cuda-10.1/lib64:$LD_LIBRARY_PATH
export PATH=$PATH:/usr/local/cuda-10.1/bin

Then I try to add cuDNN libraries. However I found two CUDA folders under /use/local:

cuda
cuda-10.1

I run nvcc -V in both folders and they are both version 10.1. So now I have two questions:

  1. Should I copy cuDNN libraries to cuda/include or cuda-10.1/include or both?

  2. Why did I get two folders? Seems they contain the exact same files. Should I remove one of them to make things clean?

talonmies
  • 70,661
  • 34
  • 192
  • 269
Dadi Gao
  • 31
  • 1
  • 4

1 Answers1

10

I am quite sure /usr/local/cuda is a symbolic link to /usr/local/cuda-10.1. You can check this by doing a ls -l /usr/local/cuda

Copying files to include folders in cuda/include or cuda-10.1/include should be one and the same thing.

cplusplusrat
  • 1,435
  • 12
  • 27
  • 2
    Thank for your kind explanation! You're totally right that `cuda` is the symbolic link of `cuda-10.1`. Should have done a throughout check before asking. – Dadi Gao Mar 01 '19 at 23:10
  • 1
    While both are same, which of the 2 is to be used preferably? Are there any considerations to be made? – Chaitanya Bapat Sep 06 '20 at 02:35