cuDNN 7.5.0.56 installation with local rpm under CentOS 7

AI & Data Science Deep Learning (Training & Inference) cuDNN
1

Hi Guys,
I installed CUDA 10.1 successfully, which can be checked by

[xxx cuda]$ nvcc -V
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2019 NVIDIA Corporation
Built on Fri_Feb__8_19:08:17_PST_2019
Cuda compilation tools, release 10.1, V10.1.105

Then I download the rpms of cnDNN 7.5 from the website and installed them locally with

rpm -ivh libcudnn7-*.x86_64.rpm
rpm -ivh libcudnn7-devel-*.x86_64.rpm
rpm -ivh libcudnn7-doc-*.x86_64.rpm

However I can’t find cudnn.h under

/usr/local/cuda/lib64

and libcudnn.so file also.
I run the Test of sample_common but got following error:

[xxx mnistCUDNN]$ sudo make clean && make
rm -rf *o
rm -rf mnistCUDNN
Makefile:154: ../samples_common.mk: No such file or directory
CUDA VERSION:
TARGET ARCH: x86_64
HOST_ARCH: x86_64
TARGET OS: linux
SMS:
>>> WARNING - no SM architectures have been specified - waiving sample <<<
make: *** No rule to make target `../samples_common.mk'.  Stop.

OS version is CentOS 7.3 x86_64.
I don’t know whether cnDNN has been installed succeesfully, if not, how to solve it?

2

had the same problem, the installation guide is misleading
try to copy the whole folder with samples

cp -r /usr/src/cudnn_samples_v7/ ./
then run make

3

I have the same problem. Therefore, I choose an older version, cudnn 7.4.2.
And it works.

4

Same here. I downloaded the corresponding .deb file (libcudnn7-doc_7.5.1.10-1+cuda10.1_amd64.deb), extracted the samples_common.mk and copied this to the ./cudnn_samples_v7 folder.
build and test were then sucessful.

5

You can see empty CUDA VERSION in the message.

This is because CUDA VERSION is achieved by samples_common.mk, which is outside of mnistCUDNN.

Shortly, you have to copy the whole cudnn_samples_v*, not cudnn_samples_v*/mnistCUDNN alone.