Deepin 15.8 1080ti cuda9.0 cudnn7.4
按照安裝的正確過程重新整理了一下:Deepin 15.8 肯定沒問題。
被DeepIn Linux 的操作便利性吸引了，重新構建一個學習和工作用的平台，相比Ubuntu來說，UI確實符合中文習慣，最關鍵的QQ和微信我不用折騰的，順便記錄一下過程，給後來的使用者一點參考。
Deepin linux 15.8 1080ti cuda9.0 cudnn7.4 tensorflow-gpu1.12,pytorch-gpu
1、更新補丁安裝包
- 控制中心 更新即可或者2、3步
- sudo apt update
- sudo apt upgrade 即可
2、顯卡驅動問題
cuda9.0默認帶384，不要指望，裝上去啟動不了，主要lightDM進不去，系統自帶 390 儘管符合要求，但後期不能正常工作。
處理方法：安裝cuda9.2自帶的驅動，也就是下載cuda9.2的安裝包，只安裝nv驅動，其他不安裝，我的在ubuntu18.04哪裡講了，參見該文章
先卸載自帶的驅動，有點暴力哦。呵呵，執行的過程中應該是一路的沒有，保險起見還是刪除以下，一面要蛾子。
sudo apt purge nvidia*
這樣也行
sudo apt remove --purge nvidia*
禁用開源的驅動
sudo gedit /etc/modprobe.d/blacklist-nouveau.conf
在文件中添加如下內容：
blacklist nouveau
options nouveau modeset=0
或者
執行如下命令
sudo gedit /etc/modprobe.d/blacklist.conf
在文件尾部添加：
blacklist nouveau
保存文件
執行如下命令：
sudo update-initramfs -u
先不要重啟啟動，重啟後，桌面進不去了，記住完成上面動作後一定不要重啟。。。。
3、下載cuda9.2的安裝包，不是安裝9.2，而是為了安裝它自帶的驅動
下載地址：https://developer.nvidia.com/cuda-92-download-archive?target_os=Windows&target_arch=x86_64&target_version=10
下載的文件名稱是cuda_9.2.148_396.37_linux.run，拷貝到您用戶的工作目錄，為什麼，為了不用切換路徑了，因為重啟後你要打命令的，漢字不一定可以輸入，而我們下載後通常是放在下載目錄，目錄的名字也是漢字，不拷貝，很麻煩。
4、安裝CUDA9.2 帶的顯卡驅動。
下載完成後重啟系統，桌面進不去了，不要害怕。
按CRTL+ALT+F2組合鍵，進入TTY2，登錄，在命令行輸入：
sudo chmod +x cuda_9.2.148_396.37_linux.run
sudo ./cuda_9.2.148_396.37_linux.run
或者
sudo sh ./cuda_9.2.148_396.37_linux.run
接著協議，空格翻完：
Do you accept the previously read EULA?
accept/decline/quit:accept
You are attempting to install on an unsupported configuration. Do you wish to continue?
(y)es/(n)o [ default is no ]: yes
Install NVIDIA Accelerated Graphics Driver for Linux-x86_64 396.37?
(y)es/(n)o/(q)uit: y
Do you want to install the OpenGL libraries?
(y)es/(n)o/(q)uit [ default is yes ]:
Do you want to run nvidia-xconfig?
This will update the system X configuration file so that the NVIDIA X driver
is used. The pre-existing X configuration file will be backed up.
This option should not be used on systems that require a custom
X configuration, such as systems with multiple GPU vendors.
(y)es/(n)o/(q)uit [ default is no ]: y
nstall the CUDA 9.2 Toolkit?
(y)es/(n)o/(q)uit: n
Install the CUDA 9.2 Samples?
(y)es/(n)o/(q)uit: n
等待顯卡安裝成功，我反覆多次均可以安裝成功。重啟
5、在安裝cuda9.0之前建議先安裝如下輔助文件：
sudo apt install exfat-fuse exfat-utils unrar p7zip-full p7zip-rar rar unzip gtk2-engines-pixbuf libxml2-utils build-essential gcc-4.8 gcc-4.8-multilib g++-4.8 g++-4.8-multilib gcc-5 gcc-5-multilib g++-5 g++-5-multilib gcc-6 gcc-6-multilib g++-6 g++-6-multilib gcc-7 gcc-7-multilib g++-7 g++-7-multilib curl terminator libx11-dev libxmu-dev libxi-dev libgl1-mesa-glx libglu1-mesa libglu1-mesa-dev libgles2-mesa-dev libfreeimage3 libfreeimage-dev freeglut3 freeglut3-dev libffi-dev -y
GCC G++設置 因為cuda９.0 安裝需要gcc5的編譯器，我為了方便，因為有可能其他的軟體會需要其他高版本的，我順便把gcc都給裝上，需要的時候通過配置來激活即可。
需要GCC5
當然為了方便可以考慮把所有的GGC都裝進去，看你的洗好
sudo update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-4.8 40
sudo update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-5 50
sudo update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-6 60
sudo update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-7 70
sudo update-alternatives --config gcc
會出現個選擇界面 選擇gcc5那個
sudo update-alternatives --install /usr/bin/g++ g++ /usr/bin/g++-4.8 40
sudo update-alternatives --install /usr/bin/g++ g++ /usr/bin/g++-5 50
sudo update-alternatives --install /usr/bin/g++ g++ /usr/bin/g++-6 60
sudo update-alternatives --install /usr/bin/g++ g++ /usr/bin/g++-7 70
sudo update-alternatives --config g++
選g++-5的那個
6、安裝cuda9.0
https://developer.nvidia.com/cuda-90-download-archive?target_os=Linux&target_arch=x86_64&target_distro=Ubuntu&target_version=1604&target_type=runfilelocal
下載
全部下載
sudo sh cuda_9.0.176_384.81_linux.run
Do you accept the previously read EULA?
accept/decline/quit:accept
You are attempting to install on an unsupported configuration. Do you wish to continue?
(y)es/(n)o [ default is no ]: yes
Install NVIDIA Accelerated Graphics Driver for Linux-x86_64 384.81?
(y)es/(n)o/(q)uit: no
Install the CUDA 9.0 Toolkit?
(y)es/(n)o/(q)uit: y
Enter Toolkit Location
[ default is /usr/local/cuda-9.0 ]:
Do you want to install a symbolic link at /usr/local/cuda?
(y)es/(n)o/(q)uit: y
Install the CUDA 9.0 Samples?
(y)es/(n)o/(q)uit: y
Enter CUDA Samples Location
[ default is /home/oliver ]:
進入安裝過程，
接著安裝四個補丁，同意協議，會問你安裝路徑是不是，不是改，是的回車，四個補丁一樣的。接著編輯配置文件。這個和Ubuntu18.04有點區別，我第一次把路徑信息寫到
/etc/profile文件里，重啟，居然不起任何作用，不知道是不是我個人的原因，之後我把路徑配置方法，我自己用用戶的.bashrc中，起作用了，我特地用 echo $PATH命令看了，不寫在自己用戶的.bashrc中確實不起作用，一臉懵懂。
sudo gedit ~/.bashrc
輸入如下信息，當然，你看到這個文檔的時候，拷貝也行，題外一句，我順便把Java也配置，不使用Java的請忽略
export PATH=/usr/local/cuda/bin:$PATH
export LD_LIBRARY_PATH=.:/usr/local/cuda/extras/CUPTI/lib64:/usr/local/cuda/lib64:$LD_LIBRARY_PATH
export CUDA_HOME=/usr/local/cuda
export JAVA_HOME=/usr/local/jdk8
export CLASSPATH=.:$JAVA_HOME/lib/dt.jar:$JAVA_HOME/lib/tools.jar
export PATH=$JAVA_HOME/bin:$PATH
source ~/.bashrc
或者重啟
7、接著安裝cudnn7.4
https://developer.nvidia.com/rdp/cudnn-download
下載對應版本
cudnn-9.0-linux-x64-v7.4.1.5 大約347MB
剪切到你的當前用戶的工作目錄，解壓有兩種方式：一、直接用歸檔器打開，直接拉出來，二、使用命令：
tar -xzvf cudnn-9.0-linux-x64-v7.4.1.5.tgz
sudo cp cuda/include/cudnn.h /usr/local/cuda/include
sudo cp cuda/lib64/libcudnn* /usr/local/cuda/lib64
sudo chmod a+r /usr/local/cuda/include/cudnn.h /usr/local/cuda/lib64/libcudnn*
暴力測試cuda
開啟Terminal
cd /home/oliver/NVIDIA_CUDA-9.0_Samples
make >>ll.log
等著編譯完成
cd /home/oliver/NVIDIA_CUDA-9.0_Samples/bin/x86_64/linux/release
oliver@oliver-PC:~/NVIDIA_CUDA-9.0_Samples/bin/x86_64/linux/release$ ./convolutionFFT2D
[./convolutionFFT2D] - Starting...
GPU Device 0: "GeForce GTX 1080 Ti" with compute capability 6.1
Testing built-in R2C / C2R FFT-based convolution
...allocating memory
...generating random input data
...creating R2C & C2R FFT plans for 2048 x 2048
...uploading to GPU and padding convolution kernel and input data
...transforming convolution kernel
...running GPU FFT convolution: 4400.440143 MPix/s (0.909000 ms)
...reading back GPU convolution results
...running reference CPU convolution
...comparing the results: rel L2 = 1.058234E-07 (max delta = 9.978815E-07)
L2norm Error OK
...shutting down
Testing custom R2C / C2R FFT-based convolution
...allocating memory
...generating random input data
...creating C2C FFT plan for 2048 x 1024
...uploading to GPU and padding convolution kernel and input data
...transforming convolution kernel
...running GPU FFT convolution: 4454.342975 MPix/s (0.898000 ms)
...reading back GPU FFT results
...running reference CPU convolution
...comparing the results: rel L2 = 1.087169E-07 (max delta = 1.084381E-06)
L2norm Error OK
...shutting down
Testing updated custom R2C / C2R FFT-based convolution
...allocating memory
...generating random input data
...creating C2C FFT plan for 2048 x 1024
...uploading to GPU and padding convolution kernel and input data
...transforming convolution kernel
...running GPU FFT convolution: 5547.850087 MPix/s (0.721000 ms)
...reading back GPU FFT results
...running reference CPU convolution
...comparing the results: rel L2 = 1.084221E-07 (max delta = 9.397971E-07)
L2norm Error OK
...shutting down
Test Summary: 0 errors
Test passed
oliver@oliver-PC:~/NVIDIA_CUDA-9.0_Samples/bin/x86_64/linux/release$ ./clock_nvrtc
CUDA Clock sample
> Using CUDA Device [0]: GeForce GTX 1080 Ti
> GPU Device has SM 6.1 compute capability
Average clocks/block = 3174.812500
8、Anaconda安裝
https://www.anaconda.com/download/#linux
目前最新版是python3.7，但是tensorflow目前好像3.7兼容的版本沒出，建議各位安全起見下載3.6的版本。5.2的版本帶的3.6的建議下載
安裝很簡單
chmod +x anaconda.run
./anaconda.run
9、安裝tensorflow
目前最新版是可以兼容3.7了，但安裝需要個技巧。參見解決Python3.7不能安裝tensorflow的問題，感謝作者吳子寒，如果安裝的是3.6的python 安裝很簡單：
pip install tensorflow-gpu
安裝的應該是1.12版的。速度慢，請離線下載。
pip install tensorflow_gpu-1.12.0-cp36-cp36m-manylinux1_x86_64.whl
10、安裝keras
pip install keras
11、安裝pytorch
PyTorch
因為使用anaconda 建議命令改成
pip install torch torchvision
會自動根據平台特性選擇GPU版本的pytorch,目前新的穩定版是1.0，速度慢，請離線安裝。
pip install torch-1.0.0-cp36-cp36m-manylinux1_x86_64.whl
pip install torchvision
12、安裝OpenAi 的gym
openai/gym
13、安裝AllenNLP
allenai/allennlp
其實直接
pip install allennlp
即可
14、 pycharm 的安裝參見 我ubuntu18.04 那篇文章
15、 測試（pytorch）
import torch
print(torch.__version__)
print(torch.cuda.is_available())
x=torch.randn(10)
if torch.cuda.is_available():
device=torch.device("cuda")
y=torch.ones_like(x,device=device)
x=x.to(device)
z=x+y
print(z)
print(z.to("cpu",torch.double))
結果
/home/oliver/anaconda3/bin/python /home/oliver/PycharmProjects/untitled/test.py
0.4.1.post2
True
tensor([ 1.7476, 2.9535, 1.6507, 0.2742, 0.6676, 2.6718, 2.4384, 3.2783,
-0.0905, 1.4600], device=cuda:0)
tensor([ 1.7476, 2.9535, 1.6507, 0.2742, 0.6676, 2.6718, 2.4384, 3.2783,
-0.0905, 1.4600], dtype=torch.float64)
測試沒問題
16、 tensorflow-gpu測試
import tensorflow as tf
sess = tf.Session(config=tf.ConfigProto(log_device_placement=True))
/home/oliver/anaconda3/bin/python /home/oliver/PycharmProjects/untitled/r.py
/home/oliver/anaconda3/lib/python3.7/importlib/_bootstrap.py:219: RuntimeWarning: compiletime version 3.6 of module tensorflow.python.framework.fast_tensor_util does not match runtime version 3.7
return f(*args, **kwds)
2018-12-02 12:03:41.443411: I tensorflow/core/platform/cpu_feature_guard.cc:141] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2 FMA
2018-12-02 12:03:41.538054: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:964] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2018-12-02 12:03:41.538381: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1432] Found device 0 with properties:
name: GeForce GTX 1080 Ti major: 6 minor: 1 memoryClockRate(GHz): 1.645
pciBusID: 0000:01:00.0
totalMemory: 10.91GiB freeMemory: 10.42GiB
2018-12-02 12:03:41.538392: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1511] Adding visible gpu devices: 0
Device mapping:
/job:localhost/replica:0/task:0/device:XLA_GPU:0 -> device: XLA_GPU device
/job:localhost/replica:0/task:0/device:XLA_CPU:0 -> device: XLA_CPU device
/job:localhost/replica:0/task:0/device:GPU:0 -> device: 0, name: GeForce GTX 1080 Ti, pci bus id: 0000:01:00.0, compute capability: 6.1
2018-12-02 12:03:41.694077: I tensorflow/core/common_runtime/gpu/gpu_device.cc:982] Device interconnect StreamExecutor with strength 1 edge matrix:
2018-12-02 12:03:41.694095: I tensorflow/core/common_runtime/gpu/gpu_device.cc:988] 0
2018-12-02 12:03:41.694100: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1001] 0: N
2018-12-02 12:03:41.694377: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1115] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:0 with 10078 MB memory) -> physical GPU (device: 0, name: GeForce GTX 1080 Ti, pci bus id: 0000:01:00.0, compute capability: 6.1)
2018-12-02 12:03:41.694879: I tensorflow/core/common_runtime/direct_session.cc:307] Device mapping:
/job:localhost/replica:0/task:0/device:XLA_GPU:0 -> device: XLA_GPU device
/job:localhost/replica:0/task:0/device:XLA_CPU:0 -> device: XLA_CPU device
/job:localhost/replica:0/task:0/device:GPU:0 -> device: 0, name: GeForce GTX 1080 Ti, pci bus id: 0000:01:00.0, compute capability: 6.1
Process finished with exit code 0
推薦閱讀：
