乌班图安装Pytorch、Tensorflow Cuda环境

安装Anaconda

在网址https://www.anaconda.com/products/distribution下载安装包，我这里下载的是Anaconda3-2022.05-Linux-x86_64.sh，它对应的Python版本为3.9。

执行

bash Anaconda3-2022.05-Linux-x86_64.sh

安装VIM

sudo apt-get install vim

配置Anaconda环境变量

sudo vim /etc/profile

添加环境配置(此处的user需要替换成你自己的用户名)

export PATH=/home/user/anaconda3/bin:$PATH

使配置可用

source /etc/profile

创建Anaconda虚拟环境

conda create -n py39 python=3.9.0
source activate
source deactivate
conda activate py39

下载英伟达驱动

网址https://www.nvidia.cn/geforce/drivers/

因为我这里是Geforce RTX 3090,所以在下载页面依次填写

搜索后选择一个驱动

我这里下载的文件名为NVIDIA-Linux-x86_64-470.94.run

执行

sudo vim /etc/modprobe.d/blacklist.conf

在末尾添加

blacklist nouveau

保存后继续执行

sudo update-initramfs -u
reboot

重启后重新激活py39

conda activate py39

卸载旧版本的驱动

sudo apt-get purge nvidia*
sudo apt-get autoremove

给下载的run文件赋予可执行权限

sudo chmod 777 NVIDIA-Linux-x86_64-64-470.94.run
sudo apt-get install build-essential
sudo ./NVIDIA-Linux-x86_64-470.94.run -no-x-check -no-nouveau-check -no-opengl-files

检验英伟达显卡是否安装成功

nvidia-smi

出现如上的界面表示安装成功。

安装Cuda

11.4下载地址：https://developer.nvidia.com/cuda-11-4-0-download-archive

如果不是可以去这里找https://developer.nvidia.com/cuda-toolkit-archive

然后选择相应的选项

在终端输入

wget https://developer.download.nvidia.com/compute/cuda/11.4.0/local_installers/cuda_11.4.0_470.42.01_linux.run
chmod 777 cuda_11.4.0_470.42.01_linux.run
sudo ./cuda_11.4.0_470.42.01_linux.run --toolkit --silent --override

验证Cuda是否安装成功

cat /usr/local/cuda-11.4/version.json

如果出现

{
   "cuda" : {
      "name" : "CUDA SDK",
      "version" : "11.4.20210623"
   },
   "cuda_cudart" : {
      "name" : "CUDA Runtime (cudart)",
      "version" : "11.4.43"
   },
   "cuda_cuobjdump" : {
      "name" : "cuobjdump",
      "version" : "11.4.43"
   },
   "cuda_cupti" : {
      "name" : "CUPTI",
      "version" : "11.4.65"
   },
   "cuda_cuxxfilt" : {
      "name" : "CUDA cu++ filt",
      "version" : "11.4.43"
   },
   "cuda_demo_suite" : {
      "name" : "CUDA Demo Suite",
      "version" : "11.4.43"
   },
   "cuda_gdb" : {
      "name" : "CUDA GDB",
      "version" : "11.4.55"
   },
   "cuda_memcheck" : {
      "name" : "CUDA Memcheck",
      "version" : "11.4.43"
   },
   "cuda_nsight" : {
      "name" : "Nsight Eclipse Plugins",
      "version" : "11.4.43"
   },
   "cuda_nvcc" : {
      "name" : "CUDA NVCC",
      "version" : "11.4.48"
   },
   "cuda_nvdisasm" : {
      "name" : "CUDA nvdisasm",
      "version" : "11.4.43"
   },
   "cuda_nvml_dev" : {
      "name" : "CUDA NVML Headers",
      "version" : "11.4.43"
   },
   "cuda_nvprof" : {
      "name" : "CUDA nvprof",
      "version" : "11.4.43"
   },
   "cuda_nvprune" : {
      "name" : "CUDA nvprune",
      "version" : "11.4.43"
   },
   "cuda_nvrtc" : {
      "name" : "CUDA NVRTC",
      "version" : "11.4.50"
   },
   "cuda_nvtx" : {
      "name" : "CUDA NVTX",
      "version" : "11.4.43"
   },
   "cuda_nvvp" : {
      "name" : "CUDA NVVP",
      "version" : "11.4.43"
   },
   "cuda_samples" : {
      "name" : "CUDA Samples",
      "version" : "11.4.43"
   },
   "cuda_sanitizer_api" : {
      "name" : "CUDA Compute Sanitizer API",
      "version" : "11.4.54"
   },
   "cuda_thrust" : {
      "name" : "CUDA Thrust",
      "version" : "11.4.43"
   },
   "libcublas" : {
      "name" : "CUDA cuBLAS",
      "version" : "11.5.2.43"
   },
   "libcufft" : {
      "name" : "CUDA cuFFT",
      "version" : "10.5.0.43"
   },
   "libcurand" : {
      "name" : "CUDA cuRAND",
      "version" : "10.2.5.43"
   },
   "libcusolver" : {
      "name" : "CUDA cuSOLVER",
      "version" : "11.2.0.43"
   },
   "libcusparse" : {
      "name" : "CUDA cuSPARSE",
      "version" : "11.6.0.43"
   },
   "libnpp" : {
      "name" : "CUDA NPP",
      "version" : "11.4.0.33"
   },
   "libnvjpeg" : {
      "name" : "CUDA nvJPEG",
      "version" : "11.5.1.43"
   },
   "nsight_compute" : {
      "name" : "Nsight Compute",
      "version" : "2021.2.0.15"
   },
   "nsight_systems" : {
      "name" : "Nsight Systems",
      "version" : "2021.2.4.12"
   },
   "nvidia_driver" : {
      "name" : "NVIDIA Linux Driver",
      "version" : "470.42.01"
   }
}

表示安装成功

配置环境变量

sudo vim /etc/profile

添加内容如下

export PATH=/usr/local/cuda-11.4/bin:$PATH
export LD_LIBRARY_PATH=/usr/local/cuda-11.4/Lib64:$LD_LIBRARY_PATH
export LD_LIBRARY_PATH=/usr/local/cuda-11.4/extras/CUPTI/lib64:$LD_LIBRARY_PATH

使配置可用

source /etc/profile

查看Cuda信息

nvcc -V

显示内容如下

nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2021 NVIDIA Corporation
Built on Wed_Jun__2_19:15:15_PDT_2021
Cuda compilation tools, release 11.4, V11.4.48
Build cuda_11.4.r11.4/compiler.30033411_0

安装cuDNN

 安装cuDNN需要英伟达会员登陆，进入页面https://developer.nvidia.com/rdp/cudnn-download

这里我们下载的是Local Installer for Linux x86_64 (Tar)

解压缩

xz -d cudnn-linux-x86_64-8.4.0.27_cuda11.6-archive.tar.xz
tar xvf cudnn-linux-x86_64-8.4.0.27_cuda11.6-archive.tar

将cuDNN的文件拷贝到Cuda目录下

sudo cp cudnn-linux-x86_64-8.4.0.27_cuda11.6-archive/include/cudnn.h /usr/local/cuda-11.4/include/
sudo cp cudnn-linux-x86_64-8.4.0.27_cuda11.6-archive/lib/libcudnn* /usr/local/cuda-11.4/lib64/
sudo chmod 777 /usr/local/cuda-11.4/include/cudnn.h
sudo chmod 777 /usr/local/cuda-11.4/lib64/libcudnn*

安装Pytorch cuda版本

conda activate py39
conda install pytorch torchvision torchaudio cudatoolkit=11.3 -c pytorch

安装PyCharm

进入官网https://www.jetbrains.com/pycharm/download/#section=linux

选择社区版Community

解压缩,启动PyCharm

tar xzvf pycharm-community-2022.1.1.tar.gz
cd /home/user/下载/pycharm-community-2022.1.1/bin
./pycharm.sh

安装Tensorflow cuda版本

在py39虚拟环境下

pip install tensorboardX
pip install tf-nightly-gpu

{{o.name}}

{{m.name}}