Cudnn issues with nvidia-cuda-toolkit on ubuntu 20

I’ve setup cuda with command “sudo apt install nvidia-cuda-toolkit”, I’m running on Ubuntu 20.04 and GPU GEforce GTX 1060. Actually I’ve setup this for darknet yolo that required cudnn and opencv setup for GPU compilation. I can clearly see now the cuda went to /usr/lib/cuda folder instead of /usr/local/cuda and all folders under /lib/cuda are more less empty like lib64 and include. But nvcc command is working fine and seems cuda installation succeed but when I tried built darknet it complains with cudnn library, So can any one help me please to setup cudnn with this setup? I tried moved all the include and lib64 libraries from cudnn folder to /usr/lib/cuda/lib64 and /include respectively but it didn’t help.

Thanks in advance

1 Like

Hi @saini.jitender,

Request you to check the below link for installation steps
https://docs.nvidia.com/deeplearning/sdk/cudnn-install/index.html

Thanks!

Thanks Aakanksha for replying on this, But I checked it and this installation is for CUDA older version where cuda home dir got created in /usr/local but in my case the home folder is located /usr/lib and in this version cudnn installation is not working as listed in the guide.

Hi @saini.jitender,
There might be the issue while setting PATH/LD_LIBRARY_PATH.
However, we suggest not to use the Canonical repository for CUDA (which is what you get with nvidia-cuda-toolkit).
Use CUDA as provided by NVIDIA at https://developer.nvidia.com/cuda-downloads.
That would explain why CUDA didn’t get installed to the library the cuDNN guide is expecting.

Thanks!

Okay Aakanksha, Thank you very much again for your prompt response. Let me re-install everything again from scratch, just one information I need from you.

I’m having GTX 1060 card running on Ubuntu 20.04, Which version of CUDA and CUDNN should I start with so that it will not see any version incompatibility issue. I want to run darknet YOLO framework on this.

Thanks in advance for your support.

Hi @saini.jitender,
This post might be helpful for you.

Thanks!

Thanks Aakanksha,

I managed to install CUDA 10 with cudnn 7.6 after lots of efforts and time. Now able to run darknet yolo on this .

Thanks for your support.

Jitender

1 Like