一站式安装对应显卡版本的cuda和torch（windows）

叁歲伎倆 2024-05-06 23:30 72阅读 0赞

> **目录**
> 
> 前言
> 
> **一、安装python（也可以用conda安装）**
> 
> **二、安装VS的C++环境**
> 
> **三、安装CUDA**
> 
> 2.1、检查环境
> 
> 2.2、 查看自己是否有NVIDA显卡驱动程序
> 
> 2.3、查看GPU显卡所支持的CUDA版本
> 
> 2.4、安装对应版本的CUDA安装包
> 
> 2.5、选择对应的系统
> 
> 2.6、运行安装包
> 
> 2.7、配置Cuda的环境变量
> 
> 2.8、查看是CUDA是否安装成功
> 
> 2.9、验证deviceQuery和bandwidthTest
> 
> **四、安装pytorch**
> 
> **五、运行效果**

--------------------

## 前言 ##

> 一年前，安装过cuda，觉得并不难，就没有记录。
> 
> 这次安装还算顺利，就是在找资料的时候，浪费了不少时间
> 
> 这次就记录下来，方便以后再次安装
> 
> --------------------
> 
> **总结安装程序：**
> 
> 1、安装python环境
> 
> 2、安装VS的C++环境（因为CUDA在安装时，需要VS的里面的工具包来编译。）
> 
> 3、安装cuda
> 
> 4、安装pytorch

## 一、安装python（也可以用conda安装） ##

> **直接官网下载：**[Python Releases for Windows | Python.orgThe official home of the Python Programming Language![icon-default.png?t=N7T8][icon-default.png_t_N7T8]https://www.python.org/downloads/windows/][Python Releases for Windows _ Python.orgThe official home of the Python Programming Language_icon-default.png_t_N7T8_icon-default.png_t_N7T8_https_www.python.org_downloads_windows]
> 
> 我下载的版本——蓝奏云下载（python3.9.5）：[python-3.9.5-amd64.zip - 蓝奏云][python-3.9.5-amd64.zip -]
> 
> ![df19fcccd8954154ba836c80e603c47a.png][]
> 
> ![43b94eafc18a4099a4c97ef2c667a157.png][]
> 
> ![0e870fa10de542bc99dfc6b539bb0e62.png][]
> 
> ![a8eb0ef60e7e4cc08d35fe4a16bdd42c.png][]
> 
> **进入命令行，输入python，出现这样的界面则表示成功安装**
> 
> ![7364ba1f0f8f45ed8b7e9e17a7a4cb2f.png][]
> 
> **使用conda（**Anaconda**）的话，可以看看b站的教程（这里就不赘述了）**

## **二、安装VS的C++环境** ##

> 如果你想用其他版本可以去官网下载：[Downloads & Keys - Visual Studio Subscriptions![icon-default.png?t=N7T8][icon-default.png_t_N7T8]https://my.visualstudio.com/Downloads?q=Visual%20Studio%202022][Downloads _ Keys - Visual Studio Subscriptions_icon-default.png_t_N7T8_icon-default.png_t_N7T8_https_my.visualstudio.com_Downloads_q_Visual_20Studio_202022]
> 
> 我下载的是：**社区免费版VS 2022——VisualStudioSetup.exe**
> 
> 蓝奏云下载2022社区免费版：**https://wwm.lanzout.com/i7xQY1ods8xi**   
> 密码:6epa
> 
> ![7f90bdb2f58f44d293074a76db04aa39.png][]
> 
> 运行安装程序：
> 
> ![2a6500d0ef9248198c35157cf528358a.png][]
> 
> **然后找到C++桌面开发**
> 
> ![9162dd456216495e9fcfd7a94b45c9fe.png][]
> 
> ![22590bd740ad4bce8fa95109bfa9fc3b.png][]
> 
> 建议安装到专门放软件的盘，**不要安装到系统盘C**，然后点击开始进行安装
> 
> ![8e6ce4ddca4a4f8a803a98bf26e713c5.png][]
> 
> **安装完后，电脑需要重启，重启后就可以使用了**

## **三、**安装CUDA ##

> **在安装前，可以先看看这个博客：**[理清GPU、CUDA、CUDA Toolkit、cuDNN关系以及下载安装\_cudatoolkit-CSDN博客![icon-default.png?t=N7T8][icon-default.png_t_N7T8]https://blog.csdn.net/qq\_42406643/article/details/109545766][GPU_CUDA_CUDA Toolkit_cuDNN_cudatoolkit-CSDN_icon-default.png_t_N7T8_icon-default.png_t_N7T8_https_blog.csdn.net_qq_42406643_article_details_109545766]

> ### 2.1、检查环境 ###
> 
>  *  **检查是否有NVIDA的独立显卡**，**前面的VS是否已经安装**
> 
> ### 2.2、 **查看自己是否有NVIDA显卡驱动程序** ###
> 
> **如果没有显卡的控制面板，下载安装：**[NVIDIA GeForce 驱动程序 - N 卡驱动 | NVIDIA][NVIDIA GeForce _ - N _ _ NVIDIA]
> 
> ![e0222676b73d44d4b9560105372a47fb.png][]
> 
> ### 2.3、查看GPU显卡所支持的CUDA版本 ###
> 
> 控制面板-> 帮助-> 系统信息-> 组件-> **我的是CUDA 12.3**
> 
> ![da01b7e98f544423918c8255d7fe791b.png][]
> 
> ### 2.4、安装对应版本的CUDA安装包 ###
> 
> 进入官网：[CUDA Toolkit Archive | NVIDIA Developer][CUDA Toolkit Archive _ NVIDIA Developer]
> 
> ![b3c037614e9b46c88ba0e3ba63de8a20.png][]
> 
> ### 2.5、选择对应的系统 ###
> 
> 如果用**离线版本，就有3个G**，这里我选择的是**在线安装**
> 
> ![d394d38b6f7749e89f2de0a6c2726a8d.png][]
> 
> ![88a9bff6ab3e45c18e794f302d2167a0.png][]
> 
> ### 2.6、运行安装包 ###
> 
> ![d568138248b745da8e7ad8f3aa5055f8.png][]
> 
> 自己选择路径：
> 
> ![a0d0b1bf2dd84fdf8e3980912a9665c6.png][]
> 
> ![1898cb42fd9643bea977d5b0f65fc333.png][]
> 
> **选择——自定义安装**
> 
> ![0b6e6a8319cb46d880d38a0e627a1fb9.png][]
> 
> 安装组件，全部勾选，点击下一步
> 
> ![4c2fda27b870486c9f9b30e1d53542bf.png][]
> 
> 自定义下载路径
> 
> 建议安装到专门放软件的盘，**不要安装到系统盘C**
> 
> ![6f4c104b5e6d4194b398fe74927ca3cb.png][]
> 
> 然后慢慢等
> 
> ![cf247aaa9ab542c78c5aa3e3f4cff524.png][]
> 
> 之后一直下一步就行了~
> 
> ### 2.7、配置Cuda的环境变量 ###
> 
> 安装完成后，就需要我们配置Cuda的环境变量了
> 
> **他会自己添加的变量：**
> 
> ![a521058c69974da7bf9c8edfa0e740ec.png][]
> 
> ![2f1b801c0c164169b674d22947930ca7.png][]
> 
> 先找到我们自定义安装的cuda文件夹
> 
> ![ed5756c0a03f46b8af2d001b6a0ace60.png][]
> 
> 然后**添加两个环境变量**
> 
> ![0da296b25d7444c0b01d3af50372f59a.png][]
> 
> **相当于，总共有这4个文件夹的~**
> 
> ![2466dda3a1e04ba380ef40d12900498a.png][]
> 
> ### 2.8、**查看是CUDA是否安装成功** ###
> 
> Win + R 打开cmd ，输入命令：**nvcc --version**
> 
> ![2e75c2ada4724894a9b7ef8e6f0af611.png][]
> 
> ### 2.9、验证deviceQuery和bandwidthTest ###
> 
> 在命令窗口运行文件
> 
> ![8c29292164be44ba9265d519fddfb421.png][]
> 
> ![7a8b5f3845d04a4890291590781ab634.png][]
> 
> ![236e6a01f5bd461a97d9752b7ee63ace.png][]
> 
> ok！CUDA搞定啦~

## **四、**安装pytorch ##

>  *  **我的电脑的cuda版本是12.3的，准备安装pytorch！**
>  *  **目前，官网上没有直接支持cuda 12.3的pytorch版本！**
>  *  **通过翻阅其他博客，博主说cuda是向下兼容的！**
>  *  **我就选择了CUDA 12.1**
> 
> **4.1、通过官网，选择对应版本，然后复制命令，直接下载即可！**
> 
> ![50e55a85c583459c858c0fe60baa3708.png][]
> 
> **不过在这里先pip换源**
> 
>     pip config set global.index-url https://pypi.tuna.tsinghua.edu.cn/simple
>     pip config set install.trusted-host mirrors.aliyun.com
> 
> 同时，再设置一下超时时间，毕竟这个**torch有2~3个G**（这里就不赘述了，直接放链接，如果超时了，再来设置也行~）
> 
> [Python——记录pip问题（解决下载慢、升级失败问题）\_pip下载太慢-CSDN博客![icon-default.png?t=N7T8][icon-default.png_t_N7T8]https://blog.csdn.net/Pan\_peter/article/details/129553679][Python_pip_pip_-CSDN_icon-default.png_t_N7T8_icon-default.png_t_N7T8_https_blog.csdn.net_Pan_peter_article_details_129553679]![c1144f80eece4f3b975dea663749f137.png][]
> 
> 安装完成~
> 
> ![177a2fe02b1146f48cebb850247f1882.png][]

## 五、运行效果 ##

> 我把我之前那个项目拿来，跑一下试试~
> 
> [基于YOLOv8的多端车流检测系统（用于毕设+开源）\_yolov8 rtsp-CSDN博客![icon-default.png?t=N7T8][icon-default.png_t_N7T8]https://blog.csdn.net/Pan\_peter/article/details/132048923][YOLOv8_yolov8 rtsp-CSDN_icon-default.png_t_N7T8_icon-default.png_t_N7T8_https_blog.csdn.net_Pan_peter_article_details_132048923]

> 先下载其他库
> 
> ![c54a701f4bcd425088090df3f54cf447.png][]

> 报错了：type object 'Detections' has no attribute 'from\_yolov8'
> 
> 大佬评论：由于Detection删除了from，所以需要降级，又因为前面的sv调用，所以把supervision降级为0.6.0即可解决**pip install supervision==0.6.0**
> 
> ![5da7aa3d3dcf4ad4b36d27492408f646.png][]
> 
> 感谢大佬！！！

> 运行项目：（20多帧，比俺之前只有几帧好多了，啊哈哈~）
> 
> ![40aa537abc6441458f13a48ee9187701.png][]

> 用之前的训练demo包试试：
> 
> [windows使用YOLOv8训练自己的模型（0基础保姆级教学）\_windows10使用yolov8常见问题-CSDN博客![icon-default.png?t=N7T8][icon-default.png_t_N7T8]https://blog.csdn.net/Pan\_peter/article/details/129907710][windows_YOLOv8_0_windows10_yolov8_-CSDN_icon-default.png_t_N7T8_icon-default.png_t_N7T8_https_blog.csdn.net_Pan_peter_article_details_129907710]![357192502eaf4650baee1de084773215.png][]
> 
> 也可以正常训练
> 
> --------------------
> 
> 只不过这里遇到了一个小插曲：
> 
> 报错：**RuntimeError: DataLoader worker (pid(s) 20580, 22560, 5668, 18512, 1240, 18208, 22356) exited unexpectedly**
> 
> 他搞了多线程，我直接关闭他，把训练代码改为 workers=0
> 
>     model.train(data='./data/cat.yaml', epochs=100, imgsz=640, workers=0)
> 
> ![570da642f9a445bdb8ec2a569f83a1b7.png][]

--------------------

## 参考文章： ##

> [Cuda和cuDNN安装教程(超级详细)-CSDN博客][Cuda_cuDNN_-CSDN][全网最详细的Python安装教程（Windows） - 知乎 (zhihu.com)][Python_Windows_ - _ _zhihu.com][Cuda和cuDNN安装教程(超级详细)-CSDN博客][Cuda_cuDNN_-CSDN]
> 
> [理清GPU、CUDA、CUDA Toolkit、cuDNN关系以及下载安装\_cudatoolkit-CSDN博客][GPU_CUDA_CUDA Toolkit_cuDNN_cudatoolkit-CSDN_icon-default.png_t_N7T8_icon-default.png_t_N7T8_https_blog.csdn.net_qq_42406643_article_details_109545766]
> 
> [安装支持CUDA 12的pytorch教程 - 知乎 (zhihu.com)][CUDA 12_pytorch_ - _ _zhihu.com]

[icon-default.png_t_N7T8]: https://csdnimg.cn/release/blog_editor_html/release2.3.6/ckeditor/plugins/CsdnLink/icons/icon-default.png?t=N7T8
[Python Releases for Windows _ Python.orgThe official home of the Python Programming Language_icon-default.png_t_N7T8_icon-default.png_t_N7T8_https_www.python.org_downloads_windows]: https://www.python.org/downloads/windows/
[python-3.9.5-amd64.zip -]: https://wwm.lanzout.com/iZD2R1odri6f
[df19fcccd8954154ba836c80e603c47a.png]: https://image.dandelioncloud.cn/pgy_files/images/2024/05/06/6bae8ea9582b45a8baef6ddfb3f4ee50.png
[43b94eafc18a4099a4c97ef2c667a157.png]: https://image.dandelioncloud.cn/pgy_files/images/2024/05/06/19db6aa6d0c04805ad2b32dadae1c773.png
[0e870fa10de542bc99dfc6b539bb0e62.png]: https://image.dandelioncloud.cn/pgy_files/images/2024/05/06/088445196b694cf7b0558220984cc2cd.png
[a8eb0ef60e7e4cc08d35fe4a16bdd42c.png]: https://image.dandelioncloud.cn/pgy_files/images/2024/05/06/b3275b32999740ce955db4b98504b6ad.png
[7364ba1f0f8f45ed8b7e9e17a7a4cb2f.png]: https://image.dandelioncloud.cn/pgy_files/images/2024/05/06/4b0d9f5745a940d18c0fa4ea8c5a52ab.png
[Downloads _ Keys - Visual Studio Subscriptions_icon-default.png_t_N7T8_icon-default.png_t_N7T8_https_my.visualstudio.com_Downloads_q_Visual_20Studio_202022]: https://my.visualstudio.com/Downloads?q=Visual%20Studio%202022
[7f90bdb2f58f44d293074a76db04aa39.png]: https://image.dandelioncloud.cn/pgy_files/images/2024/05/06/04d5143185a9485fbb7fe21dd2883161.png
[2a6500d0ef9248198c35157cf528358a.png]: https://image.dandelioncloud.cn/pgy_files/images/2024/05/06/fc3e9b79af5e481eb196e171a91ecdc1.png
[9162dd456216495e9fcfd7a94b45c9fe.png]: https://image.dandelioncloud.cn/pgy_files/images/2024/05/06/b53a4f8b1dd94cb3a60b2e393ea35f61.png
[22590bd740ad4bce8fa95109bfa9fc3b.png]: https://image.dandelioncloud.cn/pgy_files/images/2024/05/06/b4ed9bd480b04c91a6cbf2ea36de6a97.png
[8e6ce4ddca4a4f8a803a98bf26e713c5.png]: https://image.dandelioncloud.cn/pgy_files/images/2024/05/06/29a4f89a0541477babc27d38e02f239b.png
[GPU_CUDA_CUDA Toolkit_cuDNN_cudatoolkit-CSDN_icon-default.png_t_N7T8_icon-default.png_t_N7T8_https_blog.csdn.net_qq_42406643_article_details_109545766]: https://blog.csdn.net/qq_42406643/article/details/109545766
[NVIDIA GeForce _ - N _ _ NVIDIA]: https://www.nvidia.cn/geforce/drivers/
[e0222676b73d44d4b9560105372a47fb.png]: https://image.dandelioncloud.cn/pgy_files/images/2024/05/06/10252715956f4086b1e03c8e7587fa42.png
[da01b7e98f544423918c8255d7fe791b.png]: https://image.dandelioncloud.cn/pgy_files/images/2024/05/06/5a789eee94ca485b8d669cc24801fd7f.png
[CUDA Toolkit Archive _ NVIDIA Developer]: https://developer.nvidia.com/cuda-toolkit-archive
[b3c037614e9b46c88ba0e3ba63de8a20.png]: https://image.dandelioncloud.cn/pgy_files/images/2024/05/06/268cd46ed5a547d7b9e3b08d2935931d.png
[d394d38b6f7749e89f2de0a6c2726a8d.png]: https://image.dandelioncloud.cn/pgy_files/images/2024/05/06/f44a5b53116946fea415cc1768036ba4.png
[88a9bff6ab3e45c18e794f302d2167a0.png]: https://image.dandelioncloud.cn/pgy_files/images/2024/05/06/e56eb82175fd40a3a7f0d4a21fc769fa.png
[d568138248b745da8e7ad8f3aa5055f8.png]: https://image.dandelioncloud.cn/pgy_files/images/2024/05/06/7bcf64c8153c461c89bcaab0f5e161dc.png
[a0d0b1bf2dd84fdf8e3980912a9665c6.png]: https://image.dandelioncloud.cn/pgy_files/images/2024/05/06/c50bdb019c6e4015bb81111c81b8c1ab.png
[1898cb42fd9643bea977d5b0f65fc333.png]: https://image.dandelioncloud.cn/pgy_files/images/2024/05/06/57692e549208477f88ce54931035fb10.png
[0b6e6a8319cb46d880d38a0e627a1fb9.png]: https://image.dandelioncloud.cn/pgy_files/images/2024/05/06/8dc7e528bfee408fb64e04d30d012851.png
[4c2fda27b870486c9f9b30e1d53542bf.png]: https://image.dandelioncloud.cn/pgy_files/images/2024/05/06/19baf02d2ce54823a64c333627fa1cdb.png
[6f4c104b5e6d4194b398fe74927ca3cb.png]: https://image.dandelioncloud.cn/pgy_files/images/2024/05/06/5b4c57e5e86c48a2856221a357405ea1.png
[cf247aaa9ab542c78c5aa3e3f4cff524.png]: https://image.dandelioncloud.cn/pgy_files/images/2024/05/06/ad20f6b55058422080b4c53039c843da.png
[a521058c69974da7bf9c8edfa0e740ec.png]: https://image.dandelioncloud.cn/pgy_files/images/2024/05/06/ccb58c2aa2914157ba2fec6f35881118.png
[2f1b801c0c164169b674d22947930ca7.png]: https://image.dandelioncloud.cn/pgy_files/images/2024/05/06/d6c5d7cec2ec44fd92c05f4c220a53c8.png
[ed5756c0a03f46b8af2d001b6a0ace60.png]: https://image.dandelioncloud.cn/pgy_files/images/2024/05/06/3c8eff3d068f4ec4a45afda25581cf03.png
[0da296b25d7444c0b01d3af50372f59a.png]: https://image.dandelioncloud.cn/pgy_files/images/2024/05/06/2f0ad66533fe4ef1a90ae1a66a16d9c4.png
[2466dda3a1e04ba380ef40d12900498a.png]: https://image.dandelioncloud.cn/pgy_files/images/2024/05/06/46f98878b64b4dbdb6002dad2fe3790f.png
[2e75c2ada4724894a9b7ef8e6f0af611.png]: https://image.dandelioncloud.cn/pgy_files/images/2024/05/06/fa3bdb7e5a0e4bfca92192bfc30b7b29.png
[8c29292164be44ba9265d519fddfb421.png]: https://image.dandelioncloud.cn/pgy_files/images/2024/05/06/78c1f40a0d3243d3839c5c72e2f26c56.png
[7a8b5f3845d04a4890291590781ab634.png]: https://image.dandelioncloud.cn/pgy_files/images/2024/05/06/290b493f1afd4ad089d9dd208f26ee65.png
[236e6a01f5bd461a97d9752b7ee63ace.png]: https://image.dandelioncloud.cn/pgy_files/images/2024/05/06/5dd1cda2ffc44fb28854a29dbb7096bc.png
[50e55a85c583459c858c0fe60baa3708.png]: https://image.dandelioncloud.cn/pgy_files/images/2024/05/06/576ebfcbcfa24792a1ec896fe82e1c60.png
[Python_pip_pip_-CSDN_icon-default.png_t_N7T8_icon-default.png_t_N7T8_https_blog.csdn.net_Pan_peter_article_details_129553679]: https://blog.csdn.net/Pan_peter/article/details/129553679
[c1144f80eece4f3b975dea663749f137.png]: https://image.dandelioncloud.cn/pgy_files/images/2024/05/06/2e4999e121814c46994a5d06b665d595.png
[177a2fe02b1146f48cebb850247f1882.png]: https://image.dandelioncloud.cn/pgy_files/images/2024/05/06/b60a39b309db4c0e8cd9c5b05e72d61a.png
[YOLOv8_yolov8 rtsp-CSDN_icon-default.png_t_N7T8_icon-default.png_t_N7T8_https_blog.csdn.net_Pan_peter_article_details_132048923]: https://blog.csdn.net/Pan_peter/article/details/132048923
[c54a701f4bcd425088090df3f54cf447.png]: https://image.dandelioncloud.cn/pgy_files/images/2024/05/06/f639f7b4a5a24948b541e7eba62acb8a.png
[5da7aa3d3dcf4ad4b36d27492408f646.png]: https://image.dandelioncloud.cn/pgy_files/images/2024/05/06/44c3085a4afa429a8217b34dd2ea964c.png
[40aa537abc6441458f13a48ee9187701.png]: https://image.dandelioncloud.cn/pgy_files/images/2024/05/06/428fdde0f6db41058ab3a3c6aab5a3e4.png
[windows_YOLOv8_0_windows10_yolov8_-CSDN_icon-default.png_t_N7T8_icon-default.png_t_N7T8_https_blog.csdn.net_Pan_peter_article_details_129907710]: https://blog.csdn.net/Pan_peter/article/details/129907710
[357192502eaf4650baee1de084773215.png]: https://image.dandelioncloud.cn/pgy_files/images/2024/05/06/e56445ffecba4035bd3eaf5242e51d0d.png
[570da642f9a445bdb8ec2a569f83a1b7.png]: https://image.dandelioncloud.cn/pgy_files/images/2024/05/06/8fa6e6e6143d4d639b11c12521d92838.png
[Cuda_cuDNN_-CSDN]: https://blog.csdn.net/jhsignal/article/details/111401628
[Python_Windows_ - _ _zhihu.com]: https://zhuanlan.zhihu.com/p/344887837
[CUDA 12_pytorch_ - _ _zhihu.com]: https://zhuanlan.zhihu.com/p/646484518#:~:text=%E5%AE%98%E7%BD%91%E4%B8%8A%E6%B2%A1%E6%9C%89%E7%9B%B4%E6%8E%A5%E6%94%AF%E6%8C%81cuda%2012%E7%9A%84pytorch%E7%89%88%E6%9C%AC%EF%BC%8C%E4%BD%86%E6%98%AF%E7%BF%BB%E9%98%85%E7%A4%BE%E5%8C%BA%E4%BA%86%E8%A7%A3%E5%88%B0%EF%BC%8Ccuda%E6%98%AF%E5%90%91%E4%B8%8B%E5%85%BC%E5%AE%B9%E7%9A%84%EF%BC%8Ccuda%2012%E5%8F%AF%E4%BB%A5%E6%94%AF%E6%8C%81,%E5%AE%98%E7%BD%91%E9%87%8C%E9%9D%A2%E7%9A%84CUDA%2011.8%EF%BC%8C%E6%89%80%E4%BB%A5%E6%88%91%E4%BB%AC%E9%80%89%E6%8B%A9Conda%20%E9%80%89%E6%8B%A9CUDA11.8%20%E5%A4%8D%E5%88%B6%E4%B8%8B%E9%9D%A2%E7%9A%84%E4%BB%A3%E7%A0%81