upgrading CuDNN

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

I am working on Jetson TX2 jetpack 3.3 cuda 9.0 CuDNN 7.0.5

I am training a model in tensorflow 1.9 but the training exits because:

Loaded runtime CuDNN library: 7.0.5 but source was compiled with: 7.1.5. CuDNN library major and minor version needs to match or have higher minor version in case of CuDNN 7.0 or later version. If using a binary install, upgrade your CuDNN library. If building from sources, make sure the library loaded at runtime is compatible with the version specified during compile configuration.
Segmentation fault (core dumped)

so i tried to install CuDNN 7.3.0 for Cuda 9.0 by:

Download CuDNN library from NVIDIA website
sudo tar -xzvf [TAR_FILE]
sudo cp cuda/include/cudnn.h /usr/local/cuda/include
sudo cp cuda/lib64/libcudnn* /usr/local/cuda/lib64
sudo chmod a+r /usr/local/cuda/include/cudnn.h /usr/local/cuda/lib64/libcudnn*

in the terminal its giving me:

#define CUDNN_MAJOR 7
#define CUDNN_MINOR 3
#define CUDNN_PATCHLEVEL 0

but the error from tensorflow did not go

What did I do wrong can you guide me please?

edit:the CuDNN 7.3.0 that i installed is not for arm processors that was my fault but do you know where to get CuDNN for arm?

2

Why did you install 7.0.3? The error message says you should install 7.1.5.

3

sorry I meant 7.3.0

4

My advice would be: 1. make sure /usr/loca/cuda is pointing to the right version of cuda in case you have more than one installed. 2. try to install cuDNN 7.1.5.

Also, what was the new error from tensorflow? Did it still complain about 7.0.5? Maybe it finds cuDNN in another place on the system.

1 Like
5

thanks for your replay
1-cuda is pointing at the right version ,i only have one.
2-in the CuDNN archive (cuDNN Archive | NVIDIA Developer) there is no v 7.1.5 and the error says it accepts later versions.
3-its the same error it reads the old version 7.0.5

keep in mind i am on jetson

edit:
i found a libcudnn.so.7.0.5 in /usr/lib/aarch64-linux-gnu

6

edit:the CuDNN 7.3.0 that i installed is not for arm processors that was my fault but do you know where to get CuDNN for arm?

7

update: CuDNN arm for jetson is only installable through jetpack
for me i wanted 7.1.5 which comes with jetpack 3.3 so i flashed the jetson and all good now

8

I have the same problem, I can’t find the v7.1.5 cudnn for arm architectures, did you solved it?

9

Yes CUDNN v 7.1.5 is in jetpack 3.3 installation package, for me I flashed the jetson and worked.

10

Thank you @remi.coulom. I had two versions of CUDA installed. 10.0 and 10.2. I removed 10.0 away from 'C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA ’ directory and it worked for me.

11

system config:
os : ubuntu 18.04
GPU: 1060Ti 6gb
CUDA is 10.0,
CUDNN is 7.5
Tensorflow 1.15.2

Loaded runtime CuDNN library: 7.5.0 but source was compiled with: 7.6.0. CuDNN library major and minor version needs to match or have higher minor version in case of CuDNN 7.0 or later version. If using a binary install, upgrade your CuDNN library. If building from sources, make sure the library loaded at runtime is compatible with the version specified during compile configuration

Greatfull comments welcome,
Thanks.