Set correct path but when "make" cudnn it try to use wrong path.

Hi!

I had a problem with setting up cuda. OS is ubuntu 14.04, GPU is NVIDIA K80. I followed http://developer.download.nvidia.com/compute/cuda/7.5/Prod/docs/sidebar/CUDA_Installation_Guide_Linux.pdf
to install cuda 8.0. I did every step in pre-installation, installation, and post-installation. I set path to /usr/src/local/cuda8.0/bin and ld path accordingly and I can use nvcc command. But when I try to validate by cuda-install-example, it finally said “/usr/bin/ld: cannot find -lnvcuvid”. I continue to install cudnn following “http://docs.nvidia.com/deeplearning/sdk/cudnn-install/index.html”, I downloaded three .deb and installed. When I try to validate by mninst example, it says “/bin/sh: 1: /usr/local/cuda/bin/nvcc: not found”. Why it tried to use a path which is never set and not shown in the env? How can I fix that? Any comment is welcomed. Thanks in advance!

The -lnvcuvid issue is discussed elsewhere. Please search for it.

Your path issue may relate to your actual CUDA install path /usr/src/… vs. the one the mnist example is looking for /usr/local/…

So you’ll need to sort that out.