Don’t panic in case of data tilt, teach you to easily obtain the slope of table tilt>>> 
There are many problems in installing tensorflow GPU today
1. The download of PIP install tensorflow GPU is too slow
the solution can be found in the common Python machine learning modules
2. After installing tensorflow, importerror will appear: could not find ‘cudart64’_ 90. DLL ‘
solution: in https://developer.nvidia.com/cuda-toolkit When downloading CUDA 9.0, you must download the corresponding version and then install it.
generally, the default installation location is C:// program files/NVIDIA GPU computing toolkit/CUDA/v9.0
CUDA_ PATH_ V9_ 0 = C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v9.0
CUDA_ SDK_ PATH = C:\ProgramData\NVIDIA Corporation\CUDA Samples\v9.0
CUDA_ LIB_ PATH = %CUDA_ PATH%\lib\x64
CUDA_ BIN_ PATH = %CUDA_ PATH%\bin
CUDA_ SDK_ BIN_ PATH = %CUDA_ SDK_ PATH%\bin\win64
CUDA_ SDK_ LIB_ PATH = %CUDA_ SDK_ PATH%\common\lib\x64
3. After installing CUDA 9.0, it still appears: importerror: could not find ‘cudnn64’_ 7. DLL ‘
solution: in https://developer.nvidia.com/cudnn Then download the corresponding version, unzip the file, and add D:: (cudnn-9.2-windows10-x64-v7.1) CUDA bin to cudnn64_ 7. The DLL file is copied to C:// program files/NVIDIA GPU computing toolkit/CUDA/v9.0/bin, and it’s done.
— –
Author: Yao Xianxian
source: CSDN
original text: https://blog.csdn.net/u011311291/article/details/81393933
Copyright notice: This article is the original article of the blogger, please attach the blog link if you reprint it