cankaole
/p/a16aaa1551d2
1/注意:安装驱动
(1)检查自己电脑的gpu是否CUDA-capable
lspci | grep -i nvidia
(2)删除旧的NVIDIA驱动
sudo apt-get remove nvidia-*sudo apt-get autoremove
(3)
sudo apt-get update
(4)使用下面的命令查看系统推荐安装哪个版本的NVIDIA显卡驱动
ubuntu-drivers devices
(5)选择你看到推荐版本安装,本例使用nvidia-driver-440,然后安装几个必要组件,命令如下
sudo apt-get install nvidia-settings nvidia-driver-440 nvidia-prime
(6)
sudo reboot
(7)查看是否安装成功
nvidia-smi
2\安装低版本gcc
sudo apt install gcc-7 g++-7
将gcc 7 g++7 设置为优先级最高
sudo update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-7 90 --slave /usr/bin/g++ g++ /usr/bin/g++-7 --slave /usr/bin/gcov gcov /usr/bin/gcov-7sudo update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-9 70 --slave /usr/bin/g++ g++ /usr/bin/g++-9 --slave /usr/bin/gcov gcov /usr/bin/gcov-9
然后gcc --version
默认返回7.x版
gcc-9 --version
返会9.X版本
3\安装cuda
sudo sh cuda_10.2.89_440.33.01_linux.run
if error
sudo sh ./cuda_10.2.89_440.33.01_linux.run --toolkit --silent --override
安装时勾选去掉显卡驱动
7/zai在bashrc中添加,主目录中ctrl+h就会出来
export PATH=/usr/local/cuda-10.2/bin:$PATHexport LD_LIBRARY_PATH=/usr/local/cuda-10.2/lib64${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}
8/重新载入
source ~/.bashrc
9检查cuda
nvcc --version
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) - NVIDIA Corporation
Built on Wed_Oct_23_19:24:38_PDT_
Cuda compilation tools, release 10.2, V10.2.89
CUDNN先解压包
(1)
tar -xzvf cudnn-10.2-linux-x64-v8.2.1.32.tgz
(2)复制到cuda下
sudo cp cuda/include/cudnn.h /usr/local/cuda/includesudo cp cuda/lib64/libcudnn* /usr/local/cuda/lib64sudo chmod a+r /usr/local/cuda/include/cudnn.h /usr/local/cuda/lib64/libcudnn*
(3)安装deb包
sudo dpkg -i libcudnn8_8.2.1.32-1+cuda10.2_amd64.debsudo dpkg -i libcudnn8-dev_8.2.1.32-1+cuda10.2_amd64.debsudo dpkg -i libcudnn8-samples_8.2.1.32-1+cuda10.2_amd64.deb
(4)验证cudnn
复制cudnn实例并编译它
cp -r /usr/src/cudnn_samples_v8/ $HOMEcd $HOME/cudnn_samples_v8/mnistCUDNN
(4)
sudo apt-get install libfreeimage3 libfreeimage-devmake clean && make
输入./mnistCUDNN
返回Test passed!
,cudnn安装成功!
./mnistCUDNN
if error
sudo reboot
安装pytorch
(1)在页面上选择相应版本
Start Locally | PyTorch
pip3 install torch torchvision torchaudio
(2)重启电脑后测试
python3import torchprint(torch.cuda.is_available())