I get a Segmentation Fault error when running require 'cudnn'

There is no reason when doing the same for torch, cudnn or cutorch

My environment is:
GPU: Tesla K80
CUDA: 9.2
Lua: 5.2.3