文章目录
安装版本信息:
- NIVIDA Driver:
±----------------------------------------------------------------------------+
| NVIDIA-SMI 470.94 Driver Version: 470.94 CUDA Version: 11.4 |
|-------------------------------±---------------------±---------------------+NVIDIA-Linux-x86_64-470.94.run
- CUDA: 11.2 (Cuda compilation tools, release 11.2, V11.2.152)
cuda_11.2.2_460.32.03_linux.run
- cudnn: 8.2.1
libcudnn8_8.2.1.32-1+cuda11.3_amd64.deb
libcudnn8-dev_8.2.1.32-1+cuda11.3_amd64.deb
libcudnn8-samples_8.2.1.32-1+cuda11.3_amd64.deb
- TensorRT:
相关应用版本信息:
- Conda:
Anaconda3-5.3.1-Linux-x86_64.sh
- Pytorch:
torch==1.9.0+cu111 torchvision==0.10.0+cu111 torchaudio==0.9.0
- PaddlePaddle:
paddlepaddle-gpu==2.2.2.post112
驱动层
参考文章:
安装NVIDIA驱动
测试通过机型(成功安装无报错):
- 3060(6G)系列笔记本
- 3060(12G)系列台式机
- 3070(8G)系列台式机
安装CUDA和cudnn
将CUDA库的路径加入系统变量:(可选项)
export CPATH=/usr/local/cuda-11.2/targets/x86_64-linux/include:$CPATH
export LD_LIBRARY_PATH=/usr/local/cuda-11.2/targets/x86_64-linux/lib:$LD_LIBRARY_PATH
cudnn安装测试
- 复制
the cuDNN samples
至HOME:
cp -r /usr/src/cudnn_samples_v8/ $HOME & cd $HOME/cudnn_samples_v8/mnistCUDNN
- 编译
the mnistCUDNN sample
:
make clean && make
编译报错解决:
- 运行
the mnistCUDNN sample
:
./mnistCUDNN
- 最终结果显示CUDA相关信息,且结果为PASS:
cd /usr/local/cuda/samples/1_Utilities/deviceQuery
sudo make
./deviceQuery
安装TensorRT8
软件层
pytorch
(1)Pytorch1.9.0-cuda111
本文将采用conda进行安装,命令全部基于官网教程:
- 其他安装方式可参考官网
conda create -n totch python=3.7
conda activate torch
pip install torch==1.9.0+cu111 torchvision==0.10.0+cu111 torchaudio==0.9.0 -f https://download.pytorch.org/whl/torch_stable.html
测试:
import torch
torch.cuda.is_available()
(2)Open-mmlab-cuda111
conda create -n open-mmlab python=3.7
conda activate open-mmlab
pip install torch==1.9.0+cu111 torchvision==0.10.0+cu111 torchaudio==0.9.0 -f https://download.pytorch.org/whl/torch_stable.html
(3)mmlab相关库
- mmcv
pip install mmcv-full -f https://download.openmmlab.com/mmcv/dist/cu111/torch1.9.0/index.html
- mmdet
pip install mmdet
- 剩余步骤参考官网
PaddlePaddle
PaddlePaddle2.2.2-cuda11.2
conda create -n paddle python=3.7
conda activate paddle
python -m pip install paddlepaddle-gpu==2.2.2.post112 -f https://www.paddlepaddle.org.cn/whl/linux/mkl/avx/stable.html
# 如果上述命令运行后报错
#
测试:
import paddle
paddle.utils.run_check()
应用移除
卸载
cudnn
如果之前是以deb的方式安装的, 则:
sudo dpkg -l | grep cudnn
卸载:
sudo dpkg -r libcudnn8-samples
sudo dpkg -r libcudnn8-dev
sudo dpkg -r libcudnn8
检查:
sudo dpkg -l | grep cudnn
卸载完成.