今天装tensorflow-gpu出现了很多问题
1.pip install tensorflow-gpu下载过慢
解决办法可查看 Python机器学习常用模块
2.安装完tensorflow以后,运行时出现ImportError: Could not find ‘cudart64_90.dll’
解决办法:在https://developer.nvidia.com/cuda-toolkit上下载CUDA 9.0,一定要下载对应的版本,然后安装,
一般默认的位置安装的位置在C:Program FilesNVIDIA GPU Computing ToolkitCUDAv9.0。
安装完以后,再设置环境变量
CUDA_PATH = C:Program FilesNVIDIA GPU Computing ToolkitCUDAv9.0
CUDA_PATH_V9_0 = C:Program FilesNVIDIA GPU Computing ToolkitCUDAv9.0
CUDA_SDK_PATH = C:ProgramDataNVIDIA CorporationCUDA Samplesv9.0
CUDA_LIB_PATH = %CUDA_PATH%libx64
CUDA_BIN_PATH = %CUDA_PATH%in
CUDA_SDK_BIN_PATH = %CUDA_SDK_PATH%inwin64
CUDA_SDK_LIB_PATH = %CUDA_SDK_PATH%commonlibx64
3.安装完CUDA 9.0以后,还是出现:ImportError: Could not find ‘cudnn64_7.dll’
解决办法:在https://developer.nvidia.com/cudnn 上注册,然后下载相应版本,解压缩文件,然后将D:cudnn-9.2-windows10-x64-v7.1cudaincudnn64_7.dll 文件拷贝到C:Program FilesNVIDIA GPU Computing ToolkitCUDAv9.0in里面,就大功告成了
---------------------
作者:姚贤贤
来源:CSDN
原文:https://blog.csdn.net/u011311291/article/details/81393933
版权声明:本文为博主原创文章,转载请附上博文链接!