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:
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:
uname -m && cat /etc/*release
结果显示:
4. 确定系统已经安装了GCC
输入:
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:
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,
sudo apt-get install linux-headers-$(uname -r)
1.安装CUDA
下载CUDA,从https://developer.nvidia.com/cuda-downloads,下载对应版本的cuda安装包,我下载的是deb(local)版,
安装CUDA,运行如下命令,即可安装CUDA;
sudo dpkg -i cuda-repo-ubuntu1404_7.5-18_amd64.deb
sudo apt-get update
sudo apt-get install cuda
安装完成后,重启一下;
2.安装cuDNN
下载cuDNN,从https://developer.nvidia.com/rdp/cudnn-download,注册,然后下载,
注意事项:
有可能由于最新cuDNN不稳定,导致后续caffe工程编译失败,报出如下错误,详情参考第三个链接,这时可以回退一下,下载较新版本的cuDNN;
CXX src/caffe/test/test_im2col_layer.cpp
In file included from ./include/caffe/util/device_alternate.hpp:40:0,
from ./include/caffe/common.hpp:19,
from ./include/caffe/blob.hpp:8,
from src/caffe/test/test_im2col_layer.cpp:5:
./include/caffe/util/cudnn.hpp: In function ‘void caffe::cudnn::createPoolingDesc(cudnnPoolingStruct**, caffe::PoolingParameter_PoolMethod, cudnnPoolingMode_t*, int, int, int, int, int, int)’:
./include/caffe/util/cudnn.hpp:127:41: error: too few arguments to function ‘cudnnStatus_t cudnnSetPooling2dDescriptor(cudnnPoolingDescriptor_t, cudnnPoolingMode_t, cudnnNanPropagation_t, int, int, int, int, int, int)’
pad_h, pad_w, stride_h, stride_w));
^
./include/caffe/util/cudnn.hpp:15:28: note: in definition of macro ‘CUDNN_CHECK’
cudnnStatus_t status = condition;
^
In file included from ./include/caffe/util/cudnn.hpp:5:0,
from ./include/caffe/util/device_alternate.hpp:40,
from ./include/caffe/common.hpp:19,
from ./include/caffe/blob.hpp:8,
from src/caffe/test/test_im2col_layer.cpp:5:
/usr/local/cuda/include/cudnn.h:799:27: note: declared here
cudnnStatus_t CUDNNWINAPI cudnnSetPooling2dDescriptor(
^
Makefile:572: recipe for target '.build_release/src/caffe/test/test_im2col_layer.o' failed
make: *** [.build_release/src/caffe/test/test_im2col_layer.o] Error 1
拷贝cuDNN库文件到cuda目录下,
tar -zxvf cudnn-7.0-linux-x64-v4.0-prod.tgz
cd cuda
sudo cp lib64/* /usr/local/cuda/lib64/
sudo cp include/cudnn.h /usr/local/cuda/include/
设置环境变量,在/etc/profile中添加CUDA环境变量,
sudo gedit /etc/profile
在打开的文件中加入如下两句话
export PATH=/usr/local/cuda/bin:$PATH
export LD_LIBRARY_PATH=/usr/local/cuda/lib64:$LD_LIBRARY_PATH
保存后,使环境变量立即生效,
source /etc/profile
进入/usr/local/cuda/samples,执行下面的命令来build samples,
sudo make all -j4
全部编译完成后,进入 samples/bin/x86_64/linux/release,运行deviceQuery,
./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,
# 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,
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,
make pycaffe
make distribute
4.mnist数据测试
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;