记录本人的Cuda+Cudnn+Pytorch+TensorRT配置流程。配置过多台Ubuntu18.04/20.04系统及其深度学习环境,流程均如下且有效。
系统环境
- Ubuntu 18.04/20.04
- Python:3.8.0
- Cuda:11.3
- Cudnn:8.6.0
- Pytorch:1.12.1+cu113
- TensorRT:8.4.1.5
1. Anaconda
-
1不成功则用Index of /anaconda/archive/ | 清华大学开源软件镜像站 | Tsinghua Open Source Mirror
-
下载后bash xx.sh
-
source ~/.bashrc # 测试是否安装成功 conda # 不成功则 gedit ~/.bashrc # 添加以下内容 export PATH=/home/***/anaconda3/bin:$PATH # 创建python环境 conda create -n py38 python=3.8
2. Cuda
-
软件和更新 -> 附加驱动中安装Nvidia驱动(一般选nividia-driver-470即可),选择想安装的驱动,点击应用更改。装完重启ubuntu(给远程设备配置时,需在重启前设置用户自动登录)。
-
# ubuntu18.04直接执行以下: wget https://developer.download.nvidia.com/compute/cuda/11.3.1/local_installers/cuda_11.3.1_465.19.01_linux.run sudo sh cuda_11.3.1_465.19.01_linux.run # 其他版本: # https://developer.nvidia.com/cuda-11-3-1-download-archive?target_os=Linux&target_arch=x86_64&Distribution=Ubuntu
-
安装过程中:
- continue
- accept
- driver取消选中
- 选中install
-
gedit .bashrc # 打开后,末尾加入以下五行: # NVIDIA CUDA Toolkit export PATH=/usr/local/cuda-11.3/bin:$PATH export LD_LIBRARY_PATH=/usr/local/cuda-11.3/lib64 # 自启动conda环境 conda activate py38 # 检查是否成功 source ~/.bashrc nvcc -V
3. Cudnn
-
官网下载文件,Local Installer for Linux x86_64 (Tar)
-
tar -xf cudnn-linux-x86_64-8.6.0.163_cuda11-archive.tar.xz cd cudnn-linux-x86_64-8.6.0.163_cuda11-archive/ sudo cp -P lib/* /usr/local/cuda/lib64/ sudo cp -P include/* /usr/local/cuda/include/ # 检查是否成功 cat /usr/local/cuda/include/cudnn_version.h | grep CUDNN_MAJOR -A 2
4. Pytorch
- 可参考官网链接选择对应版本
- 也可直接如下步骤安装
# 在conda虚拟环境中安装 conda activate py38 pip3 install torch==1.12.1+cu113 torchvision==0.13.1+cu113 torchaudio==0.12.1 --extra-index-url https://download.pytorch.org/whl/cu113 # 验证 python import torch print(torch.cuda.is_available())
5. TensorRT
-
https://developer.nvidia.com/tensorrt下载对应版本的tensorrt(需要登陆nvidia账号)
# 8.4 # https://developer.nvidia.com/nvidia-tensorrt-8x-download # TensorRT 8.4 GA # TAR Package
-
解压下载的tensorrt包,得到文件夹(bin data doc graphsurgeon include lib onnx_graphsurgeon python samples targets TensorRT-Release-Notes.pdf uff)
-
#添加路径到~/.bashrc sudo gedit ~/.bashrc #加入 export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/home/***/TensorRT-8.4.1.5/lib source ~/.bashrc # 安装tensorrt cd python pip install tensorrt-8.4.1.5-cp38-none-linux_x86_64.whl # 安装uff、graphsurgeon、onnx_graphsurgeon # cd到对应文件夹下 pip install xxxx.whl # 验证 python import tensorrt