win10下pytorch和CUDA的安装完整过程

我会带着你远行 2023-07-04 12:52 5阅读 0赞

**一、安装Anaconda 或者Miniconda，这里不再赘述，conda的具体安装和使用自行百度。**

**二、查看并升级Nvidia显卡驱动**

1、在win10系统里，点击右下角的向上箭头，看到缩略图，英伟达的logo下方有个“！”，提示驱动有更新，若已是最新版本，跳过以下2、3两步。

![20200211123015139.png][]

2、点击后跳转到NVIDIA官网，点击“自动驱动程序更新”下的下载按钮；

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dlaXhpbl80MTc2MjE3Mw_size_16_color_FFFFFF_t_70][]

3、下载完成并安装后重启计算机，打开桌面上新安装的GeForce Experience程序，进入界面后，完成登录或者注册后，下载最新版本的GeForce Game Ready Driver，下载完成后安装即可，安装完成后可能会需要重启计算机；

4、查看当前显卡驱动版本，鼠标右键，打开NVIDIA控制面板，选择 “帮助”下的“系统信息”，可看到当前的显卡驱动版本，

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dlaXhpbl80MTc2MjE3Mw_size_16_color_FFFFFF_t_70 1][]

**三、安装CUDA**

1、下载正确版本的cuda:

进入[Nvidia CUDA下载官网][Nvidia CUDA]，下载对应版本的CUDA-Toolkit，首先在官网查看对应[Release Notes][]版本发布信息：

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dlaXhpbl80MTc2MjE3Mw_size_16_color_FFFFFF_t_70 2][]

由于conda的清华镜像源还没有cuda10.2的版本，这里我们可下载10.1版本, [CUDA10.1下载地址][CUDA10.1]。

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dlaXhpbl80MTc2MjE3Mw_size_16_color_FFFFFF_t_70 3][]

2、安装CUDA：

点击下载好的exe安装程序后，检查系统兼容性后选择自定义安装，选择安装以下4个组件即可，点击下一步，安装完成。

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dlaXhpbl80MTc2MjE3Mw_size_16_color_FFFFFF_t_70 4][]

如果安装完后没有报错，并且C:\\ProgramData\\NVIDIA Corporation\\CUDA Samples\\v10.1和C:\\Program Files\\NVIDIA GPU Computing Toolkit\\CUDA\\v10.1目录路径都存在的话，可进行下一步设置环境变量。

3、设置环境变量

在系统变量中添加以下变量（与Path同级位置）

CUDA_PATH = C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.1
    CUDA_PATH_V9_1 = C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.1
    
    ##正常情况下上面两个电脑直接生成了
    
    CUDA_SDK_PATH = C:\ProgramData\NVIDIA Corporation\CUDA Samples\v10.1
    CUDA_LIB_PATH = %CUDA_PATH%\lib\x64
    CUDA_BIN_PATH = %CUDA_PATH%\bin
    CUDA_SDK_BIN_PATH = %CUDA_SDK_PATH%\bin\win64
    CUDA_SDK_LIB_PATH = %CUDA_SDK_PATH%\common\lib\x64

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dlaXhpbl80MTc2MjE3Mw_size_16_color_FFFFFF_t_70 5][]

依次填入后，找到path依次点击新建,将 %CUDA\_LIB\_PATH%；%CUDA\_BIN\_PATH%；%CUDA\_SDK\_BIN\_PATH%；%CUDA\_SDK\_LIB\_PATH%；填完后如下图，全部安装好后一定要记得**重启！。**

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dlaXhpbl80MTc2MjE3Mw_size_16_color_FFFFFF_t_70 6][]

4、测试CUDA是否安装成功：

进入cuda安装路径：C:\\Program Files\\NVIDIA GPU Computing Toolkit\\CUDA\\v10.1\\extras\\demo\_suite

同样用cmd，分别输入deviceQuery.exe 和 bandwidthTest.exe，若都显示 Rsult=PASS 则说明通过，否则需要重新安装。

**四、安装pytorch**

1、conda添加清华镜像源：

由于从官方的conda源中下载速度过于缓慢，我们需要配置清华镜像源：

conda config --add channels https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/free/
    conda config --add channels https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main/
    conda config --set show_channel_urls yes

另外为了保险起见，建议同时添加第三方conda源：

conda config --add channels https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud/conda-forge/
    conda config --add channels https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud/msys2/
    conda config --add channels https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud/bioconda/
    conda config --add channels https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud/menpo/
    conda config --add channels https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud/pytorch/

2、创建并进入虚拟环境

conda create -n pytorch

conda activate pytorch

3、安装pytorch

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dlaXhpbl80MTc2MjE3Mw_size_16_color_FFFFFF_t_70 7][]

进入pytorch官网，根据Python和CUDA选择对应的版本，然后官方给出提示可通过运行：

conda install pytorch torchvision cudatoolkit=10.1 -c pytorch  
**但是这里一定要注意，去掉-c pytorch，安装的时候才会默认从清华源下载相应的包，因此这里用命令行:**

conda install pytorch torchvision cudatoolkit=10.1

接着等待安装成功就好了。

最后可以通过命令行验证一下，可以看到已经正确安装PyTorch1.4.0，且可以调用GPU。

import torch
    
    print(torch.__version__)
    print("gpu", torch.cuda.is_available())

打印结果：

![20200211140636419.png][]

**至此，pytorch以及GPU显卡的配置大功告成！**

[20200211123015139.png]: https://img-blog.csdnimg.cn/20200211123015139.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dlaXhpbl80MTc2MjE3Mw_size_16_color_FFFFFF_t_70]: https://img-blog.csdnimg.cn/20200211123449920.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dlaXhpbl80MTc2MjE3Mw==,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dlaXhpbl80MTc2MjE3Mw_size_16_color_FFFFFF_t_70 1]: https://img-blog.csdnimg.cn/20200211124501826.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dlaXhpbl80MTc2MjE3Mw==,size_16,color_FFFFFF,t_70
[Nvidia CUDA]: https://developer.nvidia.com/cuda-downloads
[Release Notes]: https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html#abstract
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dlaXhpbl80MTc2MjE3Mw_size_16_color_FFFFFF_t_70 2]: https://img-blog.csdnimg.cn/20200211125129257.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dlaXhpbl80MTc2MjE3Mw==,size_16,color_FFFFFF,t_70
[CUDA10.1]: https://developer.nvidia.com/cuda-10.1-download-archive-update1
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dlaXhpbl80MTc2MjE3Mw_size_16_color_FFFFFF_t_70 3]: https://img-blog.csdnimg.cn/20200211130151610.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dlaXhpbl80MTc2MjE3Mw==,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dlaXhpbl80MTc2MjE3Mw_size_16_color_FFFFFF_t_70 4]: https://img-blog.csdnimg.cn/20200211132004135.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dlaXhpbl80MTc2MjE3Mw==,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dlaXhpbl80MTc2MjE3Mw_size_16_color_FFFFFF_t_70 5]: https://img-blog.csdnimg.cn/20200211133009613.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dlaXhpbl80MTc2MjE3Mw==,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dlaXhpbl80MTc2MjE3Mw_size_16_color_FFFFFF_t_70 6]: https://img-blog.csdnimg.cn/20200211133059143.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dlaXhpbl80MTc2MjE3Mw==,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dlaXhpbl80MTc2MjE3Mw_size_16_color_FFFFFF_t_70 7]: https://img-blog.csdnimg.cn/20200211140131583.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dlaXhpbl80MTc2MjE3Mw==,size_16,color_FFFFFF,t_70
[20200211140636419.png]: https://img-blog.csdnimg.cn/20200211140636419.png