记录一次 ubuntu 18.04 安装CUDA 10.1+TensorFlow1.13.1的过程

首先，安装CUDA10必须用到的库

sudo apt update &&
sudo apt-get install gcc make -y &&
sudo apt-get install openjdk-8-jdk git python-dev python3-dev python-numpy python3-numpy python-six python3-six build-essential python-pip python3-pip python-virtualenv swig python-wheel python3-wheel libcurl3-dev libcupti-dev -y &&
sudo apt-get install libcurl4-openssl-dev -y

给source list添加graphic drivers

sudo add-apt-repository ppa:graphics-drivers/ppa &&
sudo apt update &&
sudo apt upgrade -y

检查一下是否挂载了gpu

sudo apt install ubuntu-drivers-common -y &&
ubuntu-drivers devices

安装最新的驱动

sudo ubuntu-drivers autoinstall

下载并安装CUDA 10.1

wget https://developer.nvidia.com/compute/cuda/10.1/Prod/local_installers/cuda-repo-ubuntu1804-10-1-local-10.1.105-418.39_1.0-1_amd64.deb

sudo dpkg -i cuda-repo-ubuntu1804-10-1-local-10.1.105-418.39_1.0-1_amd64.deb 
sudo apt-key add /var/cuda-repo-10-1-local-10.1.105-418.39/7fa2af80.pub
sudo apt-get update
sudo apt-get install cuda -y

安装cuDNN，下载下面这3个文件(这里没法使用wget下载，所以我用的方法是传到s3上，其他人可以使用其他方式复制到Linux上)

sudo apt install awscli -y
aws configure

aws s3 cp s3://mlliuz-files/cudnn/ . --recursive &&

sudo dpkg -i ./libcudnn8_8.0.5.39-1+cuda10.1_amd64.deb &&
sudo dpkg -i ./libcudnn8-dev_8.0.5.39-1+cuda10.1_amd64.deb &&
sudo dpkg -i ./libcudnn8-samples_8.0.5.39-1+cuda10.1_amd64.deb

重启

sudo reboot

安装Conda和TensorFlow 1.13.1

wget https://repo.anaconda.com/archive/Anaconda3-2020.11-Linux-x86_64.sh
chmod +x Anaconda3-2020.11-Linux-x86_64.sh
./Anaconda3-2020.11-Linux-x86_64.sh

export PATH="~/anaconda3/bin:$PATH" &&
source ~/.bashrc &&
conda create -n py36 python=3.6 -y &&
conda activate py36 &&
conda install tensorflow-gpu==1.13.1 -y 

下载测试代码，看一下跑的速度，来判断有没有成功

wget https://bastudypic.oss-cn-hongkong.aliyuncs.com/2021/01/tf.py
python tf.py