CUDA, Tensorflow-gpu 오류

1. 시스템 환경

 

OS Version : Window 10

CUDA : cuda_10.0.130_411.31_win10

CUDALib :  cudnn-10.0-windows10-x64-v7.6.4.38

Python : 3.7.7

Tensorflow-gpu : 2.1.0

 

 

2. 오류 내용

 

UnknownError:  Failed to get convolution algorithm. This is probably because cuDNN failed to initialize, so try looking to see if a warning log message was printed above.
    [[node sequential_4/sequential_1/model1/Conv1/Conv2D (defined at <ipython-input-3-d0c437e9f2e6>:28) ]] [Op:__inference_distributed_function_13526]

Function call stack:
distributed_function

 

3. 해결 방법

 

CUDA와 Tensorflow-gpu와 버전이 맞지 않는 문제였음Tensorflow-gpu 를 2.0.0 으로 downgrade하니까 해결됨

 

conda uninstall tensorflow-gpu
conda install tensorflow-gpu=2.0.0