最近又搭了一次环境(Ubuntu14.04、CUDA8.0、cuDNN5.1、Torch7、zbs_torch)。

安装好Ubuntu14.04后,首先安装CUDA8.0,其次是cuDNN5.1,之后是Torch7,最后是zbs_torch。

  • CUDA8.0:为了安装CUDA8.0,要卸载系统自带的显卡驱动

    sudo apt-get remove xserver-xorg-video*

    在官网下载到CUDA最新版的deb包(这里是cuda-repo-ubuntu1404-8-0-local_8.0.44-1_amd64.deb),然后安装CUDA(安装CUDA的过程会自动安装对应的NVIDIA驱动)

    sudo dpkg -i cuda-repo-ubuntu1404-8-0-local_8.0.44-1_amd64.deb

    sudo apt-get update

    sudo apt-get install cuda

    编译CUDA Sample

    cd /usr/local/cuda/samples

    sudo make

    编译完成后,运行deviceQuery可以显示显卡信息

    cd bin/x86_64/linux/release

    sudo ./deviceQuery

  • cuDNN5.1:在官网下载到cuDNN的两个deb包(cuDNN v5.1 Runtime Library for Ubuntu14.04 (Deb)和cuDNN v5.1 Developer Library for Ubuntu14.04 (Deb)),首先双击安装Runtime包,之后双击安装Developer包
  • Torch7

    curl -s https://raw.githubusercontent.com/torch/ezinstall/master/install-deps | bash

    git clone https://github.com/torch/distro.git ~/torch –recursive

    cd ~/torch; ./install.sh

    安装到最后输入yes把Torch的路径加入到全局设置,最后更新bashrc

    source ~/.bashrc

  • zbs_torch:首先安装luarocks

    sudo apt-get install luarocks

    之后安装mobdebug

    luarocks install mobdebug

    下载zbs_torch

    git clone https://github.com/soumith/zbs-torch.git

    启动zbs_torch

    cd zbs-torch
    ./zbstudio.sh

    开启debugger服务:”Project->Start Debugger Server”,使用下面的语句debug

    require(‘mobdebug’).start()

    设置为Torch7:”Project->Lua Interpreter->Torch-7″