安装时间:2019-03-14
显卡: 1066
环境: win10 + anaconda 4.6.8 + python 3.6.8 + CUDA 10.0 + cuDNN 7.3.1 + Tensorflow-gpu 1.13.1 3-14
环境: win10 + anaconda 4.6.8 + python 3.6.8 + CUDA 10.0 + cuDNN 7.4.1 + Tensorflow-gpu 1.13.1 更新时间3-17 安装openCV 4.0.1 更新时间3-22
下载网址:
CUDA : https://developer.nvidia.com/cuda-toolkit-archive
cuDNN历史版: https://developer.nvidia.com/rdp/cudnn-archive
cuDNN最新版: https://developer.nvidia.com/rdp/cudnn-download
安装步骤及需要注意的地方:
1.安装最新的显卡驱动;
2.安装CUDA是,注意要用自定义安装,并只勾选CUDA组件,其余都去掉(特别是最后一个会重新安装老版本的显卡驱动)
3.解压cuDNN,覆盖到Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.0
4.打开CMD,输入 nvcc -V 确认CUDA版本
5.输入conda creat --name mytf python=3.6 //建立一个命名为mytf的环境,可以自定义
6.输入activate mytf //激活自定义环境,并进入
7.输入pip install tensorflow-gpu //安装TF
8.验证环节参考其他博客
采坑史:
用了最新的CUDA 10.1 + cuDNN 7.5.0 , 失败。 2019-03-13
更新:
2019-03-17报了这么一个错,现象为显存爆炸。。根据推荐安装了 cuDNN 7.4.1,可以正常运行
2019-03-17 19:51:42.004078: E tensorflow/stream_executor/cuda/cuda_dnn.cc:324] Loaded runtime CuDNN library: 7.3.0 but source was compiled with: 7.4.1. CuDNN library major and minor version needs to match or have higher minor version in case of CuDNN 7.0 or later version. If using a binary install, upgrade your CuDNN library. If building from sources, make sure the library loaded at runtime is compatible with the version specified during compile configuration.
tensorflow.python.framework.errors_impl.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 inference/Conv2D}}]]