Loaded runtime CuDNN incompatible with version used to compile source

Hi everyone,
I have the common issue : Loaded runtime CuDNN library: 8.0.5 but source was compiled with: 8.1.0

But I cannot solve it, and more problematic I do not know where there is a loaded runtime CuDNN lib version 8.0.5 as I have installed myself the 8.1.x version and there is not other that I am aware of.

My config:


nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2021 NVIDIA Corporation
Built on Mon_May__3_19:41:42_Pacific_Daylight_Time_2021
Cuda compilation tools, release 11.3, V11.3.109
Build cuda_11.3.r11.3/compiler.29920130_0


tensorflow.version = 2.5.0
build.build_info[‘cuda_version’]) = 64_112
build.build_info[‘cudnn_version’]) = 64_8

I have installed
cudnn-11.2-windows-x64-v8.1.1.33 → same issue
Where can a version 8.0.5 can be found???

Need help now :-)
Best regards

1 Like

I can only add, that I have the exact same issue:

E tensorflow/stream_executor/cuda/cuda_dnn.cc:352] Loaded runtime CuDNN library: 8.0.5 but source was compiled with: 8.1.0. CuDNN library needs to have matching major version and equal or higher minor 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.

My libs are:
TensorFlow 2.5.0
Cuda 11.2
cuDNN 8.1.1

1 Like

Hi @j_massot , Apologies for delayed response, Are you still facing the issue.

I have the exact same error, with the same lib versions

For everyone experiencing similar issues. In my case, I had loaded torch and tensorflow simultaneously resulting in the above error as pytorch ships with its version of cudnn.