参考文献

下载网址

https://developer.nvidia.com/cuda-downloads https://developer.nvidia.com/rdp/form/cudnn-download-survey

DiskGenius分区

用DiskGenius进行分区调整和新分区,为Linux预留下空间。
https://www.diskgenius.cn/help/createpartex.phpGPU torch 默认_虚拟机

用UltraISO做USB启动盘

1.选择ultraiso,右键单击,选择“以管理员身份运行”;

2.点击左上角“文件”,选择“打开”,最后选择需要的操作系统文件,点击“打开”;

  

GPU torch 默认_docker_02

3.选择“启动”,再选择“写入硬盘映像”;

4.选择u盘,写入方式选择“USB-HDD+”,最后点击“写入”;

5.等待u盘写入操作系统镜像完成即可。

GPU torch 默认_GPU torch 默认_03


输入WIN R,然后输入diskmgmt.msc,按回车键名即可打开磁盘管理, C盘留300M, 最后一个盘留500G给Linux。这些空间必须是未分配状态。选中较快的SSD硬盘作为Linux分区,后续要进行AI计算

GPU torch 默认_虚拟机_04


SSD就是磁盘0,预留了500G

GPU torch 默认_虚拟机_05

设置BIOS USB启动

设置Secure Boot Control Disabled, USB Preboot mode Enabled

GPU torch 默认_GPU torch 默认_06

GPU torch 默认_GPU torch 默认_07

BIOS无USB启动

F2进入Bios, 没有USB启动,中文无解,查外文

• USB device is plugged in
• F2 to BIOS
• Switch to “Boot” and set “FastBoot” to Disabled
• Switch to “Security” and set “Secure Boot Control” to Disabled
• F10 to save and exit and immediately press F2 to BIOS
• Only then could I switch to “Boot” and set “Launch CSM” to Enabled
• F10 to save and exit and immediately press F2 to BIOS
• Now select a boot option. My USB option didn’t say USB, it was just [ 8.0.7]
• F10 to save and exit – booted right to USB

设置成功

GPU torch 默认_虚拟机_08

Ubuntu 安装, 选something else

GPU torch 默认_GPU torch 默认_09

安装五笔

ubuntu20.04

john@john-HP-Laptop:~/Desktop$ sudo apt-get install ibus-rime

john@john-HP-Laptop:~/Desktop$ sudo apt-get install librime-data-wubi librime-data-pinyin-simp

john@john-HP-Laptop:~$ ibus restart

GPU torch 默认_docker_10


GPU torch 默认_GPU torch 默认_11


shift直接切换五笔和拼音英文模式。

任何模式下直接回车可输入英文,但长英文只能是拼音模式。

任何模式下直接shift加回车就是大写字母。点击+号

GPU torch 默认_docker_12

再选择WuBi和Pinyin两种输入法

GPU torch 默认_windows_13


ubuntu18.04

sudo apt install freewb

GPU torch 默认_虚拟机_14


GPU torch 默认_虚拟机_15

左右手错位

左右手的切换花了一个小时解决问题

GPU torch 默认_windows_16

光标自动触发

GPU torch 默认_清除密码_17

镜像源

把Ubuntu的服务器source.list换成清华源
/etc/apt/sources.list

创建文件/home/.pip/pip.conf, 内容如下:
[global]
timeout = 6000
index-url = https://pypi.tuna.tsinghua.edu.cn/simple
trusted-host = pypi.tuna.tsinghua.edu.cn

因为许多的科学软件及其数据都在境外,但该网站限制VPN工具的介绍,关于如何安装VPN,也只好请大家自行搜索安装。

启用Ubuntu的休眠模式

sudo apt install gnome-tweak-tool

gnome-shell --version

Ubuntu 18.04下电源菜单中添加(Hibernate)休眠选项

https://ywnz.com/linuxjc/1577.html

https://extensions.gnome.org/extension/755/hibernate-status-button/

GPU torch 默认_虚拟机_18


/dev/nvme0n1p5 373108736 388732927 15624192 7.5G Linux swap

sudo systemctl suspend or the hibernate option: sudo systemctl hibernate

open /etc/default/grub

find GRUB_CMDLINE_LINUX_DEFAULT= line - these are the option added to the regular boot menu choices

add the resume=/dev/nvme0n1p5 option to the list (with correct swap partition path) like this:

Before: GRUB_CMDLINE_LINUX_DEFAULT=“nosplash enable_mtrr_cleanup=1”

After: GRUB_CMDLINE_LINUX_DEFAULT=“nosplash enable_mtrr_cleanup=1 resume=/dev/sda6”

save file

in the terminal execute command (to actually enable the new configuration settings)

sudo update-grub2

在经历以上惨痛经历后,仍未获成功

省电设置

Ubuntu通过命令睡眠、休眠
$ sudo apt-get install laptop-mode-tools  # 安装laptop-mode,否则直接用挂起命令会导致系统down掉,只能强制关机
$ cat /proc/sys/vm/laptop_mode        # 命令查看当前laptop-mode情况,若为0,则说明没开启
$ sudo gedit /etc/laptop-mode/laptop-mode.conf # 修改配置文件
修改ENABLE_LAPTOP_MODE_ON_AC为 1,即插上电源时可以用命令挂起(没插时就直接合上盖子吧)
$ sudo laptop_mode start            # 启动laptop_mode
$ cat /proc/sys/vm/laptop_mode         # 查看是否成功启动,不为0则成功;按上面的配置是2
$ systemctl suspend               # 挂起电脑,与合上盖子一个效果
$ systemctl hibernate              # 休眠,内存中内容拷入硬盘

Ubuntu 18.04下电源菜单中添加(Hibernate)休眠选项
https://ywnz.com/linuxjc/1577.html

Vitis-AI 要求

The GPU docker has been tested with GPU machines with Docker 19.03.1, NVIDIA driver 410.xx (to work with CUDA 10.0) and nvidia-docker 2.2.2.

Nvidia驱动

禁止自带的nouveau nvidia驱动
打开配置文件
sudo vim /etc/modprobe.d/blacklist-nouveau.conf
填写禁止配置的内容:
blacklist nouveau
options nouveau modeset=0
更新配置文件,
sudo update-initramfs -u

john@john-wang:~/Vitis-AI/docker$ ubuntu-drivers devices
== /sys/devices/pci0000:00/0000:00:01.0/0000:01:00.0 ==
modalias : pci:v000010DEd0000139Asv00001043sd00001A8Dbc03sc02i00
vendor : NVIDIA Corporation
model : GM107M [GeForce GTX 950M]
driver : nvidia-driver-440 - distro non-free recommended
driver : nvidia-driver-435 - distro non-free
driver : nvidia-driver-390 - distro non-free
driver : xserver-xorg-video-nouveau - distro free builtin

如果遇到上面的命令没有回显,选中下面的第三项,再更新后就有了

GPU torch 默认_虚拟机_19

依赖库安装

sudo apt-get install freeglut3-dev build-essential libx11-dev libxmu-dev libxi-dev libgl1-mesa-glx libglu1-mesa libglu1-mesa-dev

Cuda Install

Installation Instructions:
sudo apt-key add /var/cuda-repo-10-0-local-10.0.130-410.48/7fa2af80.pub
sudo dpkg -i cuda-repo-ubuntu1804-10-0-local-10.0.130-410.48_1.0-1_amd64.deb
sudo apt-get update
sudo apt-get install cuda

sudo dpkg -i cuda-repo-ubuntu1804-10-0-local-nvjpeg-update-1_1.0-1_amd64.deb

sudo vim /etc/proflie

#set nvidia cuda env

export CUDA_HOME=/usr/local/cuda

export PATH=GPU torch 默认_虚拟机_20CUDA_HOME/bin

export LD_LIBRARY_PATH=/usr/local/cuda/lib64KaTeX parse error: Expected '}', got 'EOF' at end of input: …LIBRARY_PATH:+:{LD_LIBRARY_PATH}}

GPU torch 默认_docker_21


source ~/.bashrc

sudo apt install nvidia-cuda-toolkit测试CUDA是否安装成功,终端输入命令:nvcc -V

GPU torch 默认_docker_22


cd /usr/local/cuda-10.0/samples/1_Utilities/deviceQuery

sudo make

./deviceQuery

GPU torch 默认_windows_23


若出现下面的错误,

cudaGetDeviceCount returned 30 -> unknown error

输入命令

nvidia-smi

可能出现:

failed to initialize NVML: Driver/library version mismatch

重启系统即可

安装Cudnn

https://docs.nvidia.com/deeplearning/sdk/cudnn-install/index.html

john@john-wang:~/Downloads$ sudo dpkg -i libcudnn7_7.6.5.32-1+cuda10.0_amd64.deb
john@john-wang:~/Downloads$ sudo dpkg -i libcudnn7-dev_7.6.5.32-1+cuda10.0_amd64.deb
john@john-wang:~/Downloads$ sudo dpkg -i libcudnn7-doc_7.6.5.32-1+cuda10.0_amd64.deb

验证cudnn
cudnn本身的bug
Editing /usr/include/cudnn.h, changing the line:
sudo vim /usr/include/cudnn.h
#include “driver_types.h” to #include <driver_types.h>

john@john-wang:~/Downloads$ cp -r /usr/src/cudnn_samples_v7/ $HOME
john@john-wang:~/Downloads$ cd  $HOME/cudnn_samples_v7/mnistCUDNN
john@john-wang:~/cudnn_samples_v7/mnistCUDNN$ make clean && make
john@john-wang:~/cudnn_samples_v7/mnistCUDNN$ ./mnistCUDNN

GPU torch 默认_虚拟机_24

调出五笔

GPU torch 默认_windows_25



快捷键设置

GPU torch 默认_虚拟机_26


GPU torch 默认_GPU torch 默认_27

版本选择

tensorflow版本与cuda和cudnn的对应关系:
https://tensorflow.google.cn/install/source

Anaconda安装

对于anaconda3 的安装非常简单,从官网中直接下载3.5版本的sh文件。然后执行如下命令对conda进行安装,我下载的是Anaconda3-5.1.0-Linux-x86_64.sh,过程直接yes、yes安装即可,对于不懂的可以看这个更详细的教程。
bash Anaconda3-5.1.0-Linux-x86_64.sh
安装完成后要重启电脑才能打开jupyter notebook。重启之后在终端输入一下命令进入notebook:

jupyter notebook
打开notebook界面如下,是生成在浏览器中的.
.TensorFlow和Keras 安装
安装完anaconda 以后可以在终端直接用pip 对TensorFlow和Keras进行安装:

安装 gpu 版本的 tensorflow keras opencv

sudo pip3 install tensorflow-gpu==1.14.0 # 默认安装最新版本
sudo pip3 install keras==2.2.4
sudo pip3 install -i https://pypi.tuna.tsinghua.edu.cn/simple tensorflow==1.14.0 
sudo pip3 install -i https://pypi.tuna.tsinghua.edu.cn/simple install keras==2.2.4

对于测试的代码可以使用手写数字识别进行测试,在GitHub这里可以找到。我用自己的项目跑了一下,是在jupyter notebook上运行的,使用的框架是Keras。

下面是使用keras opencv图片时的一通狂安装

pip3 install numpy

python3 -m pip install matplotlib

GPU torch 默认_GPU torch 默认_28


降级:

pip install h5py==2.10 -i https://pypi.tuna.tsinghua.edu.cn/simple/安装opencv时,如果出现下面的问题,请依照下面的顺序安装

GPU torch 默认_windows_29

pip3 install --upgrade pip
pip3 install opencv-python

保存调试镜像

docker ps -l
docker commit ID repository
docker images

进入GPU镜像

容器内输入nvidia-smi,检查是否载入GPU驱动
docker run --runtime=nvidia \

Docker内图像显示

-v /tmp/.X11-unix:/tmp/.X11-unix \ #共享本地unix端口
-e DISPLAY=$DISPLAY \ #修改环境变量DISPLAY

鼠标复位

dconf reset -f /org/mate/desktop/peripherals/mouse/

截图快捷键

GPU torch 默认_windows_30

双系统突变单系统

重新装引导区
sudo update-grub
sudo grub-install /dev/sda

锁屏

正常情况无法关闭锁屏时,用命令行
gsettings get org.gnome.desktop.lockdown disable-lock-screen
false

禁止更新

sudo gedit /etc/apt/apt.conf.d/10periodic

APT::Periodic::Update-Package-Lists "0";
APT::Periodic::Download-Upgradeable-Packages "0";
APT::Periodic::AutocleanInterval "0";
APT::Periodic::Unattended-Upgrade "0";

sudo gedit /etc/apt/apt.conf.d/20auto-upgrades

APT::Periodic::Update-Package-Lists "0";
APT::Periodic::Download-Upgradeable-Packages "0";
APT::Periodic::AutocleanInterval "0";
APT::Periodic::Unattended-Upgrade "0";

**

TWEAK TOOL

**
【sudo add-apt-repository universe】
回车再次键入命令:【sudo apt install gnome-tweak-tool】