只信真爱 發表於 2020-4-30 20:16:00

Ubuntu 安装 CUDA

<p>显卡：Nvidia GF MX150</p>
<p>CUDA：CUDA(ComputeUnified Device Architecture)，是显卡厂商NVIDIA推出的运算平台。 CUDA是一种由NVIDIA推出的通用并行计算架构，该架构使GPU能够解决复杂的计算问题。</p>
<p>CuDNN：NVIDIA cuDNN是用于深度神经网络的GPU加速库。它强调性能、易用性和低内存开销。NVIDIA cuDNN可以集成到更高级别的机器学习框架中，如谷歌的Tensorflow、加州大学伯克利分校的流行caffe软件。简单的<strong>插入式设计</strong>可以让开发人员专注于设计和实现神经网络模型，而不是简单调整性能，同时还可以在GPU上实现高性能现代并行计算。</p>
<h3>1、安装Nvidia驱动</h3>
<p>快捷安装，打开软件和更新-&gt;附加驱动，自动搜索显卡驱动，选择其中一个驱动版本；点击应用更改等待安装完成，重启即可；（如下图）</p>
<p><img style="display: block; margin-left: auto; margin-right: auto" src="https://img2020.cnblogs.com/blog/1365039/202004/1365039-20200430194525967-1057048749.png" alt="" width="513" height="369"></p>
<h3>2、安装CUDA10.2</h3>
<p>利用deb安装快捷简便</p>
<p>（1）打开CUDA官网，找到相应的版本（如下图）[网址]，按照说明安装；</p>
<p><img style="display: block; margin-left: auto; margin-right: auto" src="https://img2020.cnblogs.com/blog/1365039/202004/1365039-20200430194558139-565167333.png" alt="" width="904" height="521"></p>
<p>（2）配置CUDA环境变量</p>
<p>在.bashrc末尾添加两行环境变量</p>
<div class="cnblogs_code">
<pre>export PATH=$PATH:$/usr/local/cuda-<span style="color: rgba(128, 0, 128, 1)">10.2</span>/<span style="color: rgba(0, 0, 0, 1)">bin#根据CUDA版本更换路径
export LD_LIBRARY_PATH</span>=/usr/local/cuda-<span style="color: rgba(128, 0, 128, 1)">10.2</span>/lib64${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}#根据CUDA版本更换路径</pre>
</div>
<p>保存并退出；并执行</p>
<div class="cnblogs_code">
<pre>source ~/.bashrc</pre>
</div>
<p>（3）重启电脑，并测试CUDA</p>
<div class="cnblogs_code">
<pre>cd /usr/local/cuda/samples/1_Utilities/<span style="color: rgba(0, 0, 0, 1)">deviceQuery
sudo make
.</span>/deviceQuery</pre>
</div>
<p>如出现以下FAIL，则未重启电脑，重启后即会解决</p>
<div class="cnblogs_code">
<pre>./<span style="color: rgba(0, 0, 0, 1)">deviceQuery Starting...

CUDA Device Query (Runtime API) version (CUDART </span><span style="color: rgba(0, 0, 255, 1)">static</span><span style="color: rgba(0, 0, 0, 1)"> linking)

cudaGetDeviceCount returned </span><span style="color: rgba(128, 0, 128, 1)">803</span>
-&gt; system has unsupported display driver /<span style="color: rgba(0, 0, 0, 1)"> cuda driver combination
Result </span>= FAIL</pre>
</div>
<p>如出现以下，则安装成功</p>
<div class="cnblogs_code">
<pre>--------------------------------------------------<span style="color: rgba(0, 0, 0, 1)">
deviceQuery, CUDA Driver </span>= CUDART, CUDA Driver Version = <span style="color: rgba(128, 0, 128, 1)">10.2</span>, CUDA Runtime Version = <span style="color: rgba(128, 0, 128, 1)">10.2</span>, NumDevs = <span style="color: rgba(128, 0, 128, 1)">1</span><span style="color: rgba(0, 0, 0, 1)">
Result </span>= PASS</pre>
</div>
<h3>3、安装CuDNN</h3>
<p>（1）下载[网址]，需要提前注册，找到对应版本，下载cuDNN Library for Linux压缩包即可</p>
<p>（2）解压压缩包，并终端进入文件夹，执行以下</p>
<div class="cnblogs_code">
<pre>sudo cp cuda/include/cudnn.h /usr/local/cuda/include/<span style="color: rgba(0, 0, 0, 1)">
sudo cp cuda</span>/lib64/libcudnn* /usr/local/cuda/lib64/<span style="color: rgba(0, 0, 0, 1)">
sudo chmod a</span>+r /usr/local/cuda/include/<span style="color: rgba(0, 0, 0, 1)">cudnn.h
sudo chmod a</span>+r /usr/local/cuda/lib64/libcudnn*</pre>
</div>
<p>（3）测试</p>
<div class="cnblogs_code">
<pre>cat /usr/local/cuda/include/cudnn.h | grep CUDNN_MAJOR -A <span style="color: rgba(128, 0, 128, 1)">2</span></pre>
</div>
<p>出现以下内容，则安装成功</p>
<div class="cnblogs_code">
<pre><span style="color: rgba(0, 0, 255, 1)">#define</span> CUDNN_MAJOR 7
<span style="color: rgba(0, 0, 255, 1)">#define</span> CUDNN_MINOR 6
<span style="color: rgba(0, 0, 255, 1)">#define</span> CUDNN_PATCHLEVEL 5
--
<span style="color: rgba(0, 0, 255, 1)">#define</span> CUDNN_VERSION (CUDNN_MAJOR * 1000 + CUDNN_MINOR * 100 + CUDNN_PATCHLEVEL)<span style="color: rgba(0, 0, 0, 1)">

#include </span><span style="color: rgba(128, 0, 0, 1)">"</span><span style="color: rgba(128, 0, 0, 1)">driver_types.h</span><span style="color: rgba(128, 0, 0, 1)">"</span></pre>
</div><br><br>
来源：https://www.cnblogs.com/haijian/p/12810876.html
頁: [1]
查看完整版本: Ubuntu 安装 CUDA