Tensorflow 2.5 exit code -1073740791 when GPU training-CodePudding

When training a Tensorflow model on GPU, calling:

model.fit(...)

exits with code -1073740791 (0xC0000409) immediately:

Epoch 1/500
2021-10-16 20:13:42.154951: I tensorflow/stream_executor/platform/default/dso_loader.cc:53] Successfully opened dynamic library cudnn64_8.dll

Process finished with exit code -1073740791 (0xC0000409)

Doing the same thing on CPU works without any problem.

System information:

Python version: 3.8
CUDA version: 11.2
cuDNN version: 8.1
Tensorflow version: 2.5
OS: Windows 10 Pro

CodePudding user response：

Running the same script from terminal shows a different error, indicating that the file

cudnn_adv_infer64_8.dll

can't be found in the path.

Moving the rest of the files from the cuDNN bin folder, namely:

cudnn_adv_infer64_8.dll
cudnn_adv_train64_8.dll
cudnn_cnn_infer64_8.dll
cudnn_cnn_train64_8.dll
cudnn_ops_infer64_8.dll
cudnn_ops_train64_8.dll

into the CUDA bin folder resolved the issue.