In the same CONDA environment, install pytroch first and then tensorflow

environmental information ：

windows10

NIVDIA CUDA：11.6

python: 3.9

CUDNN：8.3

I installed it in the above environment first pytorch_1.11.0 + cuda_11.3.

Later, due to the demand, the installation continued tensorflow_gpu-2.9.0：

On the official website https://www.tensorflow.org/install/pip#windows Follow the instructions to download Python 3.9 GPU support Installation package https://storage.googleapis.com/tensorflow/windows/gpu/tensorflow_gpu-2.9.0-cp39-cp39-win_amd64.whl, utilize pip install tensorflow_gpu-2.9.0-cp39-cp39-win_amd64.whl On the installation .

Execute test statement after loading

python -c "import tensorflow as tf; print(tf.config.list_physical_devices('GPU'))"

Report errors Could not load dynamic library 'cudnn64_8.dll'; dlerror: cudnn64_8.dll not found：

W tensorflow/stream_executor/platform/default/dso_loader.cc:64] Could not load dynamic library 'cudnn64_8.dll'; dlerror: cudnn64_8.dll not found

W tensorflow/core/common_runtime/gpu/gpu_device.cc:1850] Cannot dlopen some GPU libraries. Please make sure the missing libraries mentioned above are installed properly if you would like to use GPU. Follow the guide at https://www.tensorflow.org/install/gpu for how to download and setup the required libraries for your platform.

Skipping registering GPU devices...

Error reason ： Before loading CUDNN, Maybe the new installation package is exe Packages instead of compressed packages （ It may also be that there is no compressed package ）, Put... Directly bin、lib、include Installed in C:\Program Files\NVIDIA\CUDNN\v8.3 Next , Didn't put CUDNN Of bin、lib、include The files in the directory are moved to CUDA In the corresponding directory .

No problem after moving .