Caffe学习笔记2--Ubuntu 14.04 64bit 安装Caffe（GPU版本）

106 阅读 0 评论 70 点赞

我是靠谱客的博主背后小蝴蝶，这篇文章主要介绍Caffe学习笔记2--Ubuntu 14.04 64bit 安装Caffe（GPU版本），现在分享给大家，希望可以做个参考。

0.检查配置

1. VMWare上运行的Ubuntu，并不能支持真实的GPU（除了特定版本的VMWare和特定的GPU，要求条件严格，所以我在VMWare上搭建好了Caffe环境后，又重新在Windows 7 64bit系统上安装了Ubuntu 14.04 64bit系统，链接在此，以此来搭建Caffe GPU版本）；

2. 确定GPU支持CUDA

输入：

复制代码

1
lspci | grep -i nvidia

显示结果：

我的是GTX 650，然后到http://developer.nvidia.com/cuda-gpus去验证，支持CUDA；

3. 确定Linux版本支持CUDA

输入：

复制代码

1
uname -m && cat /etc/*release

结果显示：

4. 确定系统已经安装了GCC

输入：

复制代码

1
gcc --version

结果显示：

5. 确定系统已经安装了正确的Kernel Headers和开发包

输入：

复制代码

1
uname -r

结果：4.2.0-36-generic，这个是必须安装的kernel headers和开发包的版本；

安装对应的kernels header和开发包，

复制代码

1
sudo apt-get install linux-headers-$(uname -r)

1.安装CUDA

下载CUDA，从https://developer.nvidia.com/cuda-downloads，下载对应版本的cuda安装包，我下载的是deb（local）版，

安装CUDA，运行如下命令，即可安装CUDA；

复制代码

1
2
3
sudo dpkg -i cuda-repo-ubuntu1404_7.5-18_amd64.deb
sudo apt-get update
sudo apt-get install cuda

安装完成后，重启一下；

2.安装cuDNN

下载cuDNN，从https://developer.nvidia.com/rdp/cudnn-download，注册，然后下载，

注意事项：
有可能由于最新cuDNN不稳定，导致后续caffe工程编译失败，报出如下错误，详情参考第三个链接，这时可以回退一下，下载较新版本的cuDNN；

复制代码

CXX src/caffe/test/test_im2col_layer.cpp
In file included from ./include/caffe/util/device_alternate.hpp:40:0,
from ./include/caffe/common.hpp:19,
from ./include/caffe/blob.hpp:8,
from src/caffe/test/test_im2col_layer.cpp:5:
./include/caffe/util/cudnn.hpp: In function ‘void caffe::cudnn::createPoolingDesc(cudnnPoolingStruct**, caffe::PoolingParameter_PoolMethod, cudnnPoolingMode_t*, int, int, int, int, int, int)’:
./include/caffe/util/cudnn.hpp:127:41: error: too few arguments to function ‘cudnnStatus_t cudnnSetPooling2dDescriptor(cudnnPoolingDescriptor_t, cudnnPoolingMode_t, cudnnNanPropagation_t, int, int, int, int, int, int)’
pad_h, pad_w, stride_h, stride_w));
^
./include/caffe/util/cudnn.hpp:15:28: note: in definition of macro ‘CUDNN_CHECK’
cudnnStatus_t status = condition; 
^
In file included from ./include/caffe/util/cudnn.hpp:5:0,
from ./include/caffe/util/device_alternate.hpp:40,
from ./include/caffe/common.hpp:19,
from ./include/caffe/blob.hpp:8,
from src/caffe/test/test_im2col_layer.cpp:5:
/usr/local/cuda/include/cudnn.h:799:27: note: declared here
cudnnStatus_t CUDNNWINAPI cudnnSetPooling2dDescriptor(
^
Makefile:572: recipe for target '.build_release/src/caffe/test/test_im2col_layer.o' failed
make: *** [.build_release/src/caffe/test/test_im2col_layer.o] Error 1

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
CXX src/caffe/test/test_im2col_layer.cpp
In file included from ./include/caffe/util/device_alternate.hpp:40:0,
from ./include/caffe/common.hpp:19,
from ./include/caffe/blob.hpp:8,
from src/caffe/test/test_im2col_layer.cpp:5:
./include/caffe/util/cudnn.hpp: In function ‘void caffe::cudnn::createPoolingDesc(cudnnPoolingStruct**, caffe::PoolingParameter_PoolMethod, cudnnPoolingMode_t*, int, int, int, int, int, int)’:
./include/caffe/util/cudnn.hpp:127:41: error: too few arguments to function ‘cudnnStatus_t cudnnSetPooling2dDescriptor(cudnnPoolingDescriptor_t, cudnnPoolingMode_t, cudnnNanPropagation_t, int, int, int, int, int, int)’
pad_h, pad_w, stride_h, stride_w));
^
./include/caffe/util/cudnn.hpp:15:28: note: in definition of macro ‘CUDNN_CHECK’
cudnnStatus_t status = condition; 
^
In file included from ./include/caffe/util/cudnn.hpp:5:0,
from ./include/caffe/util/device_alternate.hpp:40,
from ./include/caffe/common.hpp:19,
from ./include/caffe/blob.hpp:8,
from src/caffe/test/test_im2col_layer.cpp:5:
/usr/local/cuda/include/cudnn.h:799:27: note: declared here
cudnnStatus_t CUDNNWINAPI cudnnSetPooling2dDescriptor(
^
Makefile:572: recipe for target '.build_release/src/caffe/test/test_im2col_layer.o' failed
make: *** [.build_release/src/caffe/test/test_im2col_layer.o] Error 1

拷贝cuDNN库文件到cuda目录下，

复制代码

1
2
3
4
tar -zxvf cudnn-7.0-linux-x64-v4.0-prod.tgz
cd cuda
sudo cp lib64/* /usr/local/cuda/lib64/
sudo cp include/cudnn.h /usr/local/cuda/include/

设置环境变量，在/etc/profile中添加CUDA环境变量，

复制代码

1
sudo gedit /etc/profile

在打开的文件中加入如下两句话

复制代码

1
2
export PATH=/usr/local/cuda/bin:$PATH
export LD_LIBRARY_PATH=/usr/local/cuda/lib64:$LD_LIBRARY_PATH

保存后，使环境变量立即生效，

复制代码

1
source /etc/profile

进入/usr/local/cuda/samples，执行下面的命令来build samples，

复制代码

1
sudo make all -j4

全部编译完成后，进入 samples/bin/x86_64/linux/release，运行deviceQuery，

复制代码

1
./deviceQuery

如果出现显卡信息，则驱动及显卡安装成功，结果如下：

3.编译caffe

修改caffe/Makefile.config，

复制代码

1
2
3
4
# cuDNN acceleration switch (uncomment to build with cuDNN).
USE_CUDNN := 1#去掉这个注释
# CPU-only switch (uncomment to build without GPU support).
# CPU_ONLY := 1#加上这个注释

然后输入，

复制代码

1
2
3
4
make clean#第一次编译，不需要执行
make all
make test
make runtest

显示结果：

cpu版本Caffe学习笔记1--Ubuntu 14.04 64bit caffe安装的编测试行结果是共1058项，GPU版本的测试运行结果是2005项；

编译pycaffe，

复制代码

1
2
make pycaffe
make distribute

4.mnist数据测试

复制代码

1
2
3
4
5
cd caffe
# 下载mnist数据
sh data/mnist/get_mnist.sh
sh examples/mnist/create_mnist.sh
./examples/mnist/train_lenet.sh

执行结果，

和CPU版本Caffe学习笔记1--Ubuntu 14.04 64bit caffe安装的mnist数据测试相比，速度明显提升；当时我运行脚本后，就开始看书，刚看完一页，抬头发现已经运行完毕，第一次用GPU计算，虽然只是很渣的GTX 650，但依然对GPU的运算能力感到佩服；

5.深度学习开源库环境搭建大礼包

昨晚群里一个哥们分享了一个github链接，涵盖了主流深度学习开源库的环境搭建，包括Nvidia驱动、CUDA、cuDNN、TensorFlow、Caffe、Theano、Keras、Torch，链接在此，他的配置是Ubuntu 14.04 64bit + Nvidia Titan X，还是适用于好多朋友的机器的，所以好东西还是要分享给大家；