tf1.13.1 and dependent and associated tf2.0.0 version
Hardware Description: Graphics NVIDIA-GEFORCE-GTX-1060
1. Check the driver version and update your graphics driver [This step is very important, lower your driver version, cuda and cudnn it may be wrong]
Error: DLL load failed: The specified module could not be found.
cmd input: nvidia-smi
Graphics driver is already up to date. If not the latest video drivers, need to manually update it.
Added: drive version corresponds to the version of cuda and cudnn
We installed cuda is 10.0.130 So: nvidia driver version of win10 corresponding to greater than 411.31. [Graphics driver after I updated to: 436.48 satisfy the conditions]
If your version is lower than the 411.31 driver, you need to update the driver
Driver Download: https: //www.nvidia.com/Download/index.aspx lang = en-us?
Select the download version: choose according to their own computer graphics model (Baidu relevant models, you can have more information; can also be viewed in hardware;)
First download the standard version, suggesting that my computer windows Drivers Type for the DCH instead of the standard version, it is re-download the installation was successful.
Download drivers: 436.48-desktop-win10-64bit-international-dch-whql.exe follow the prompts (prompts) can be installed.
Description: I first uninstall the installed NVIDIA drivers after the original, if you do not uninstall the old drivers, the installation is not clear whether covering feasible.
2. Install tensorflow-gpu1.13.1 If there were no anaconda installation, install their own]
2.1 to create a virtual environment tf113:
Run as administrator in cmd: conda create -n tf113 python = 3.6.9
2.1.1 activated tf113 installation cudatoolkit and cudnn
Check for installed version: conda search cudatoolkit
Many say the Internet is not supported by 10.1, 10.1 times anyway, pre-installation of their own, without success. Here it is installed directly 10.0.130
conda install cudatoolkit = 10.0.130
Similarly: View cudnn version: conda search cudnn
Support cuda10.0 of cudnn There are two versions to choose a
conda install cudnn=7.3.1
View tensorflow-gpu alternative version
conda search tensorflow-gpu
Because the contents of a recent study, largely completed tensorflow1.13.1 basis, so here choose to install this version
conda install tensorflow-gpu=1.13.1
Now installed numpy scikit-leran Keras the like usually need to use a packet
conda install numpy
Reminder: the virtual environment, it is best to manually install the update packages, do batch updates
2.2 to create a virtual environment tf200:
2.2.1 Run as administrator in cmd: conda create -n tf200 python = 3.6.2
2.2.2: Activating a virtual environment tf200: conda activate tf200
2.2.3: Installation cuda: conda install cudatoolkit = 10.0.130
2.2.4: Installation cudnn: conda install cudnn = 7.3.1
2.2.5: Installation tensorflow-gpu 2.0.0
Note: conda search tensorflow-gpu 2.0.0 library and not the version of the package tf, so choose to install a pip
2.2.5.1: pip update to the latest version, this step is critical, not the latest version, it may not find the package tf2.0
python -m pip install --upgrade pip ## update to the latest, non-uniform way
2.2.5.2: install from source Qinghua (Tsinghua source changes :)
pip install tensorflow-gpu==2.0.0 -i https://pypi.tuna.tsinghua.edu.cn/simple