TensorFlow installation steps

环境：Win10,Python 3.8,Anaconda
电脑：GeForce RTX 2060, NVIDIA CUDA 11.1.4
官方教程：在 Windows 上安装 TensorFlow

1. 下载安装cuda和cudnn

1.1 下载

在官网上下载对应的cuda和cudnn,The version can be lower but not higher than the version supported by the computer
cuda下载地址：cuda-toolkit-archive
cudnn下载地址：cudnn-archive

常见问题： 下载cudnn的时候有提示：NVIDIA Developer Program Membership Required
解决方法： 右键选择【复制链接地址】,Copy the link to Thunder Downloader or other download software to download

1.2 安装

CUDA安装：选自定义安装→默认安装路径→安装结束
安装完成后,Two system variables are generated by default.
View system variables：此电脑→右键选属性→高级系统设置→环境变量→系统变量

CuDNN安装
解压→将三个文件夹复制到cuda的安装目录下,Just choose to overwrite the file.

安装完了,测试cuda版本.
打开cmd,输入指令：nvcc -V

2. 创建 tensorflow 环境

2.1 创建 tensorflow 环境

conda create -n tensorflow python=3.8

常见问题
ERROR：Collecting package metadata (current——repodata.json): failed
我的解决方法：将C:\Users\Lenovo下的.condarc文件作如下修改（Modify your path here）

ssl_verify: true
channels:
  - http://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/free/win-64/
  - http://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main/win-64
show_channel_urls: true

2.2 查看环境,You can see that the environment has been created**

conda env list

2.3 进入环境

activate tensorflow

enter the specific path

# Replace here with your own specific path
>cd C:\Users\Lenovo\anaconda3\envs\tensorflow

2.4 安装默认版本的tensorflow-cpu或者tensorflow-gpu.

安装tensorflow-cpu版本的,可以输入命令：pip install --ignore-installed --upgrade tensorflow
安装tensorflow-gpu版本的,可以输入命令：pip install --ignore-installed --upgrade tensorflow-gpu

常见问题
1 网速问题
解决方法：Go directly to the website to download the required files（Download whatever file your computer downloads,我的是tensorflow_gpu-2.9.1-cp38-cp38-win_amd64.whl）
网址 tensorflow-gpu 2.9.1 ,Download the file to your owntensorflow的目录下.
输入命令：pip install tensorflow_gpu-2.9.1-cp38-cp38-win_amd64.whl

2 ERROR: Could not find a version that satisfies the requirement XXX
解决方法：直接选用pip源并且信任它的来源就可以解决这种问题

pip install tensorflow_gpu-2.9.1-cp38-cp38-win_amd64.whl -i http://pypi.douban.com/simple/ --trusted-host pypi.douban.com

2.5 查看 tensorflow Version information and exit

查看 tensorflow 版本
输入命令：pip show tensorflow-gpu,可以查看tensorflow的版本信息

退出 tensorflow 环境

conda deactivate tensorflow

3. 测试 TensorFlow-gpu 是否安装成功

打开Anaconda,选择tensorflow环境,打开spyder,It needs to be installed for the first timeSpyder,直接点下方的install即可.

测试代码

import tensorflow as tf
a = tf.constant(1.)
b = tf.constant(2.)
print(a+b)
print(tf.__version__)
print(tf.test.gpu_device_name())
print('GPU:',tf.config.list_physical_devices(device_type='GPU'))
print('CPU:',tf.config.list_physical_devices(device_type='CPU'))
print(tf.test.is_gpu_available())

常见问题
Could not load dynamic library ‘cusolver64_11.dll‘; dlerror: cusolver64_11.dll not found
解决方法：
链接：https://pan.baidu.com/s/1W9fR2N_hoVD-7_ODtOiJhg
提取码：u65i
下载文件,把文件cusolver64_11.dlladded to the created environment\Library\bin中

运行代码

安装完成

3. 测试速度

import tensorflow as tf
import timeit
#指定在cpu上运行
def cpu_run():
    with tf.device('/cpu:0'):
        cpu_a = tf.random.normal([10000, 1000])
        cpu_b = tf.random.normal([1000, 2000])
        cpu_c = tf.matmul(cpu_a, cpu_b)
        # print( "cpu_a: ", cpu_a.device)
        # print( "cpu_b: ", cpu_b.device)
        # print("cpu_c:", cpu_c.device)
    return cpu_c

#指定在gpu上运行

def gpu_run():
    with tf.device( '/gpu:0'):
        gpu_a = tf.random.normal([ 10000,1000])
        gpu_b = tf.random.normal([ 1000, 2000])
        gpu_c = tf.matmul(gpu_a, gpu_b)
        # print( "gpu_a: ", gpu_a.device)
        # print("gpu_b: ", gpu_b.device)
        # print("gpu_c: ", gpu_c.device)
    return gpu_c

cpu_time = timeit.timeit(cpu_run, number = 10)
gpu_time = timeit.timeit(gpu_run, number = 10)
print('cpu:',cpu_time, 'gpu:',gpu_time)

The speed is still a lot faster

参考来源：
tensorflow详细安装教程（Win10, Anaconda,Python3.9）
绕过NvidiaOfficially required registration or login steps to download directlycuDNN的方法
详解Anaconda安装tensorflow报错问题解决方法
anaconda3There was a problem creating the virtual environment：Collecting package metadata (current——repodata.json): failed