# 前言

[En]

Recently changed a new computer, the old version of a lot of problems, a variety of errors, version mismatch problems, take a lot of detours, after a few days of trial and error, here a detailed environment for in-depth learning tutorials for your reference!

## 1. 环境介绍

windows10系统

cuda11.3.1+cudnn8.2.1，tensorflow2.7.0+Keras2.7.0，torch-1.11.0+cu113。

[En]

Two successful versions of my installation are given for your reference.

python3.8.0+CUDA11.6.0+cuDNN8.3.2+tensorflow2.7.0+Keras2.7.0
python3.8.0+CUDA11.3.1+cuDNN8.2.1+tensorflow2.7.0+Keras2.7.0

## 2. 显卡及算计要求

NVIDIA官方 CUDA与显卡驱动版本对应表查询

# ; cuda配置

## 1. 下载及安装

cuda官网网址：https://developer.nvidia.com/cuda-toolkit-archive

[En]

For the first time, it is recommended that you choose “Custom installation”. If you have installed it many times, you can choose simplified installation.

[En]

Then keep clicking on the next step, and the installation is successful.

[En]

Then keep clicking on the next step, and the installation is successful.

## ; 2. 搭建环境及测试

[En]

After the installation is successful, you need to add the appropriate configuration file to the environment variable. Find the advanced system settings in Settings-about-and click on the environment variable.

，点击【新建】，添加相应的变量名（手动添加）和变量值（就是你的CUDA的安装路径）。

# cudnn配置

cuda11.3.1+cudnn8.2.1

## 1. 下载

cuDNN官网地址：https://developer.nvidia.com/rdp/cudnn-archive，这里直接进入到版本选择界面

[En]

## ; 2. 配置环境及测试

[En]

Click OK after the configuration is successful

[En]

The following tests whether the installation is successful

[En]

The output is as follows:

# pytorch配置

[En]

Click enter and select the following requirements:

pip3 install torch torchvision torchaudio –extra-index-url

[En]

After the installation is successful, you can test whether the installation is successful by entering the following code:

import torch
from torch import nn
import torchvision


[En]

If you do not report an error, the installation is successful!

print(torch.__version__)
print(torch.version.cuda)
print(torch.backends.cudnn.version)
print(torch.cuda.get_device_name(0))
print(torch.cuda.is_available())


[En]

The running results show that:

1.11.0+cu113
11.3
8213
NVIDIA GeForce RTX 3070 Laptop GPU
True


Original: https://blog.csdn.net/zzjcymbq/article/details/125040993
Author: 勋章DhR
Title: 【超详细】windows10系统下深度学习环境搭建CUDA11.3+cuDNN，以及tensorflow，Keras，pyTorch对应版本

(0)