Ubuntu上CUDA环境搭建

0.好像装CUDA会自动装nvidia驱动，所以可以尝试直接装CUDA

 

0.1 可以尝试通过sudo apt install nvidia-cuda-toolkit来安装CUDA

 

1.下载CUDA:https://developer.nvidia.com/cuda-toolkit-archive (如果已经安装了N卡驱动，最好用.deb，如果没有安装，可以用.run)

2.根据官网提示安装CUDA

添加配置：

export PATH=$PATH:/usr/local/cuda/bin
export LD_LIBRARY_PATH=/usr/local/cuda/lib64:$LD_LIBRARY_PATH
export CUDA_ROOT=/usr/local/cuda/
export LD_LIBRARY_PATH=/opt/cudnn/lib64:$LD_LIBRARY_PATH
export LIBRARY_PATH=$LIBRARY_PATH:/opt/cudnn/lib64
export CPATH=$CPATH:/opt/cudnn/include

 

　　

3.下载Cudnn: https://developer.nvidia.com/rdp/cudnn-download

4.执行sudo dpkg -i libcudnn7-dev_7.0.3.11-1+cuda9.0_amd64.deb （更多安装方法：https://developer.nvidia.com/compute/machine-learning/cudnn/secure/v7.0.5/prod/Doc/cuDNN-Developer-Guide）

 

安装Nvidia驱动：http://www.cnblogs.com/huangshiyu13/p/8970602.html

训练时设置GPU个数：https://www.cnblogs.com/huangshiyu13/p/6479916.html