Install the CPU version of tensorflow+cuda+cudnn (ultra detailed)

  Purpose ： install GPU Version of tensorflow

One 、 Check out the NVIDIA Does the version support CUDA And configurable CUDA edition

Method ：

Right click in a blank space on the desktop , Open NVIDIA control panel , If you can't find it, you can refer to the following website  win10 nvidia How to open the control panel ,win10 nvidia Control panel opening steps - Chinese cabbage u Disk start (dabaicai.com)

  After opening , Click on the bottom left corner “ system information ”, Click again “ Components ”, notice NVCUDA.DLL Version shown later

You can see that my computer supports CUDA Of , And the highest version supports 11.4.94（ Downward compatible ）, At this time, I actually download <=11.4.94 Of CUDA All versions are available , Because downward compatibility . But you can also match it with the official website ：

First check my graphics card version ：

  Go again CUDA Check the version correspondence on the official website ： Release Notes :: CUDA Toolkit Documentation

  Then you can see that my graphics card version is 471.41, So I can download CUDA11.4 update4 Following CUDA.（ Backwards compatible ） If I want to download the above version , Then I can try to update my graphics card driver （ If it is already the latest , There's nothing we can do about it ） Update driver : Official Drivers | NVIDIA

Additional description ： I'm really Xiaobai , At that time, I didn't know how to find this version comparison when searching by myself , Even if you enter the official website, you can't find , Although many people have given links , But I still want to find it by myself , So the search process is attached below

1. Enter CUDA Official website , Then select the document （ Because I'm looking for version comparison , So I'll check the document and try ）

 2. Get into NVIDIA CUDA Tooklkit(CUDA tool kit )

 3,

  thus , Finally found

Two 、 Confirm your installation tensorflow type

For example, I want to install tensorflow_gpu_1.15.0

Then tensorflow Official website stay Windows Build from source code in the environment  |  TensorFlow

Check the supported CUDA and CUDNN Version of

 

3、 ... and 、 install CUDA

Please see what you need TENSORFLOW model , And the environment it needs ！

Because I want to install tensorflow_gpu_1.15.0, It supports CUDA The version is 7.4,CUDNN The version is 10, So I need to CUDA Download on 7.4 Version of , The method is as follows ：

Enter... In search engine CUDA+ Version number

 

  When the download is complete , Open setup , Choose where you want to install , Can be installed

Then agree and continue

 

  Note here that if the current version is higher than the new version , Then uncheck , Otherwise installation will fail

Configure environment variables

Right click on my computer stand alone --> attribute

 

  Check whether there are two environment variables in the figure below , There is no need to match , Because I have , So I don't know how to solve it , Therefore, if there is no partner, you need to search separately

thus CUDA10.0 It is installed , Verify that the installation was successful ：

First step ： Press... On the keyboard at the same time “windows key +R”, Input “cmd” And return , Get into windows Command line interface .

The second step ： Input from the command line “nvcc -V” And return

The third step ： If installed successfully CUDA Words , Will be displayed CUDA Version number of .

Four 、 install CUDNN

CUDNN download ：cuDNN Archive | NVIDIA Developer

Because I want to download tensorflow adapter 7.4 Version of CUDNN, So I download 7.4.1 Version of CUDNN

When the download is complete , Unzip downloaded cuDNN Compressed package , Copy the files to CUDA In the folder

At this time, if you want to check your CUDNN Is the installation successful , Then you can get to The following table of contents , Find out if there is cudnn.h file , If there is , It means your cudnn Installation is successful

C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.0\include

If you want to continue to check the version you installed , Open the above cudnn.h file ,  Find the following code , On behalf of you cudnn The version is 7.4.1

5、 ... and 、 install anaconda

I have installed on my computer before anaconda, Therefore, the installation process will not be put here , Partners in need need also need to use their small hands to search .

6、 ... and 、 use ANACONDA install TENSORFLOW

1） stay ANACONDA Create a file named TENSORFLOW Environment （ You can call him any name , Here I call this environment TENSORFLOW.

conda create -n tensorflow pip python=3.6

here pip python=3.6 In the name of tensorflow The build version in the environment of is 3.6 Of python.

2） Activate the environment

conda activate tensorflow 

（ To deactivate the environment and delete the environment, use the following two lines of code ）

conda deactivate
conda env remove -n tensorflow

3） install tensorflow-gpu 1.15 edition

pip install --user tensorflow-gpu==1.15 -i https://pypi.tuna.tsinghua.edu.cn/simple

The display shows that the interface instructions are installed

4） The detection is successful

Enter python, Then enter the following code to test

import tensorflow as tf
print('GPU', tf.test.is_gpu_available())

If the following interface is displayed, the installation is successful

 

  Now it's done ！

Thank you, blogger ：

Win10 install CUDA+cuDNN+TensorFlow-GPU Firefly

WIN10 install TENSORFLOW（GPU edition ） Detailed explanation （ Hyperdetail , Starting from scratch ） - You know

Anaconda Lower installation tensorflow1.15 Process record - You know