N卡驱动安装

-- nvidia支持cuda的gpu列表
https://developer.nvidia.com/cuda-gpus

-- cuda安装指南
https://docs.nvidia.com/cuda/cuda-installation-guide-linux/index.html

yum  -y install gcc gcc-c++ make cmake glibc python-devel wget

-- dkms下载
wget http://rpmfind.net/linux/fedora-secondary/releases/27/Everything/aarch64/os/Packages/d/dkms-2.3-6.20170523git8c3065c.fc27.noarch.rpm

-- 安装dkms
rpm -ivh dkms*


-- 注释自带驱动
vi /usr/lib/modprobe.d/dist-blacklist.conf
添加blacklist nouveau，
注释掉blacklist nvidiafb

mv /boot/initramfs-$(uname -r).img /boot/initramfs-$(uname -r).img.bak
dracut /boot/initramfs-$(uname -r).img $(uname -r)

-- 重启
reboot

进入字符界面
init 3

yum -y install gcc kernel-devel "kernel-devel-uname-r == $(uname -r)" dkms "kernel-devel-uname-r == $(uname -r)"

-- nvidia驱动安装
./NVIDIA-XXXX.run --kernel-source-path=/usr/src/kernels/内核号  -k $(uname -r) --dkms -s

./NVIDIA-Linux-x86_64-390.77.run --kernel-source-path=/usr/src/kernels/3.10.0-693.11.6.el7.x86_64 -k $(uname -r) --dkms -s



内核版本不一致
http://vault.centos.org/centos/7.4.1708/os/Source/SPackages/

download rpm包解压
rpm2cpio kernel*.rpm | cpio -idmv
xz -d kernel*.tar.gz
tar xvf kernel*.tar


源码编译
1、make menuconfig
2、save即可
3、make


nvidia-smi 验证nvidia驱动安装成功


安装cuda
./cuda_9.0.176_384.81_linux.run

-- cuda环境变量配置
完成后就配置cuda环境变量，编辑~/.bashrc文件
vim ~/.bashrc

export CUDA_HOME=/usr/local/cuda
export PATH=/usr/local/cuda/bin:$PATH
export LD_LIBRARY_PATH=/usr/local/cuda/lib64:$LD_LIBRARY_PATH
export LD_LIBRARY_PATH="/usr/local/cuda/lib:${LD_LIBRARY_PATH}"

source ~/.bashrc

-- cudnn安装

tar -xvzf cudnn-9.0-linux-x64-v7.3.1.20.tgz
sudo cp -P cuda/include/cudnn.h /usr/local/cuda/include 
sudo cp -P cuda/lib64/libcudnn* /usr/local/cuda/lib64 
sudo chmod a+r /usr/local/cuda/include/cudnn.h /usr/local/cuda/lib64/libcudnn*


pip换源

[global]
index-url = https://mirrors.aliyun.com/pypi/simple


pip install tensorflow-gpu==1.7
pip install mxnet-cu90


gunicorn -c gunicorn.conf face_controller:face_app

# uwsgi -p 8 --threads 5 --http 192.168.1.62:8080 --module face_controller:face_app

watch -n 0.2 nvidia-smi


export MXNET_CUDNN_AUTOTUNE_DEFAULT=0

# CPU上的计算任务的最大线程数(默认值=1) 
export MXNET_CPU_WORKER_NTHREADS=2

# 每个GPU上，进行计算的最大线程数 (默认值=2) 
export MXNET_GPU_WORKER_NTHREADS=2