1.cuda安装
下载相应版本.run文件
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
-
根据需求选择安装内容,本文未选择安装驱动
-
安装结束
-
修改配置文件
sudo gedit ~/.bashrc
CUDA_HOME=/usr/local/cuda
export PATH=$PATH:$CUDA_HOME/bin/
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:$CUDA_HOME/lib64
- 验证
nvcc -V
2.cudnn安装
- 网址https://developer.nvidia.com/cudnn
- 下载解压之后,将cuda/include/cudnn.h文件复制到usr/local/cuda/include文件夹:
sudo cp ./cuda/include/cudnn.h /usr/local/cuda/include
- 将cuda/lib64/下所有文件复制到/usr/local/cuda/lib64文件夹中,并添加读取权限:
sudo cp ./cuda/lib64/* /usr/local/cuda/lib64
sudo chmod a+r /usr/local/cuda/include/cudnn.h /usr/local/cuda/lib64/libcudnn*
3.TensorRt安装
-
确认好对应的cuda和cudnn版本
-
目前使用cuda11.3.1,cudnn8.2.1,TensorRT8.0.1.6
-
下载网址https://developer.nvidia.com/nvidia-tensorrt-8x-download
-
解压,移动至自己指定文件夹
-
sudo gedit ~/.bashrc
#tensort
export LD_LIBRARY_PATH=/home/manager/software/TensorRT-8.0.1.6/lib:$LD_LIBRARY_PATH
export LIBRARY_PATH=/home/manager/software/TensorRT-8.0.1.6/lib:$LIBRARY_PATH
source ~/.bashrc
- 安装tensorrt-python插件
cd ./TensorRT-8.0.1.6/python
pip install tensorrt-8.0.1.6-cp39-none-linux_x86_64.whl
- 安装uff组件
cd ../uff
pip install uff-0.6.9-py2.py3-none-any.whl
4 .问题
1.TensorRT was linked against cuBLAS/cuBLAS LT 11.5.1 but loaded cuBLAS/cuBLAS LT 11.4.2
- 这是因为安装的cuda是11.3.0版本的,里面的cublas不是11.5.1,安装11.3.1版本的cuda就好了