Default Tensorflow binary does not work; outdated/incompatible cuDNN version

首先定位 conda 环境 azureml_py38_PT_TF 的定义，并检查其捆绑的 cuDNN 和 TensorFlow 版本。将它们与报告的解决方法中的版本进行比较，然后使用 python3 -c "import tensorflow as tf; print(tf.config.list_physical_devices('GPU'))" 验证环境；完成的标准是默认环境无需手动替换 cuDNN 即可加载 TensorFlow 并检测到 GPU。

The azureml_py38_PT_TF conda environment is broken and doesn't work as the current (as of 17/05/2023) as the default cuDNN binary is version 6.1, which is not compatible with any version of tensorflow.

Current work around is to manually update the cuDNN binaries as below, using the link from Nvidia

export URL="CUDNN-8.6.0-LINK-FROM-NVIDIA-WEBSITE"
# ==== DOWNLOAD CUDDN ==== 
curl $URL -o ./cudnn-linux-x86_64-8.6.0.163_cuda11-archive.tar.xz 
sudo tar -xvf ./cudnn-linux-x86_64-8.6.0.163_cuda11-archive.tar.xz
# ==== INSTALL CUDDN ==== 
sudo cp ./cudnn-*-archive/include/cudnn*.h /usr/local/cuda/include 
sudo cp -P ./cudnn-*-archive/lib/libcudnn* /usr/local/cuda/lib64 
sudo chmod a+r /usr/local/cuda/include/cudnn*.h /usr/local/cuda/lib64/libcudnn*
# ==== LINK UPDATED BINARIES ==== 
sudo ldconfig
# ==== INSTALL CONDA ENV ==== 
conda create -n "tfgpu" python=3.10 -y
conda activate tfgpu
conda install -c conda-forge cudatoolkit=11.8.0 ipykernel -y
python3 -m pip install nvidia-cudnn-cu11==8.6.0.163 tensorflow==2.12.*
mkdir -p $CONDA_PREFIX/etc/conda/activate.d
echo 'CUDNN_PATH=$(dirname $(python -c "import nvidia.cudnn;print(nvidia.cudnn.__file__)"))' >> $CONDA_PREFIX/etc/conda/activate.d/env_vars.sh
echo 'export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:$CONDA_PREFIX/lib/:$CUDNN_PATH/lib' >> $CONDA_PREFIX/etc/conda/activate.d/env_vars.sh
source $CONDA_PREFIX/etc/conda/activate.d/env_vars.sh
python3 -m ipykernel install --user --name tfgpu --display-name "Python (tf-cudnn8.6)"
# ==== VERIFY ==== 
python3 -c "import tensorflow as tf; print(tf.config.list_physical_devices('GPU'))"

Is it possible to bump the included cuDNN version to get around this problem?

(Associated SO post)