Skip to content

Nvidia CUDA and CUDNN installation for Libtorch

Jump to bottom
Ash Babu edited this page Jun 16, 2023 · 5 revisions

The correct versions of Pytorch, torchvision and cuda are needed to get the object detection/segmentation working. Take a look at the latest Pytorch or the older versions and install accordingly. The easiest option would be to install using conda. The correct cuda version should be installed. This means if you used conda install pytorch torchvision torchaudio pytorch-cuda=11.7 -c pytorch -c nvidia, then the version of cuda installed should be 11.7. Download the same from here

wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu1804/x86_64/cuda-ubuntu1804.pin
sudo mv cuda-ubuntu1804.pin /etc/apt/preferences.d/cuda-repository-pin-600
wget https://developer.download.nvidia.com/compute/cuda/11.7.0/local_installers/cuda-repo-ubuntu1804-11-7-local_11.7.0-515.43.04-1_amd64.deb
sudo dpkg -i cuda-repo-ubuntu1804-11-7-local_11.7.0-515.43.04-1_amd64.deb
sudo cp /var/cuda-repo-ubuntu1804-11-7-local/cuda-*-keyring.gpg /usr/share/keyrings/
sudo apt-get update
sudo apt-get -y install cuda  ## this could be sudo apt-get install cuda-11-7

Mandatory actions

CUDAVER=cuda-11.7
export PATH=/usr/local/$CUDAVER/bin:$PATH
export LD_LIBRARY_PATH=/usr/local/$CUDAVER/lib64:$LD_LIBRARY_PATH
export CUDA_HOME=/usr/local/$CUDAVER

The above is required so that nvcc --version shows the correct cuda installation

  • Reboot the computer.

This .deb file is installed and it should produce a /usr/local/cuda11.7/ folder. Then the CMakeLists.txt will be like

set(CMAKE_CUDA_COMPILER /usr/local/cuda-11.7/bin/nvcc)
project(object_detection LANGUAGES CXX CUDA)
enable_language(CUDA)

Download the cudnn library,
more instructions Follow the debian installation from here or the tar installation as below

The downloaded file will be like cudnn-linux-x86_64-8.x.x.x_cudaX.Y-archive.tar.xz which can be used to identify x and y

Enable the local repository.

  • sudo dpkg -i cudnn-local-repo-${OS}-8.x.x.x_1.0-1_amd64.deb

Or

  • sudo dpkg -i cudnn-local-repo-${OS}-8.x.x.x_1.0-1_arm64.deb

Import the CUDA GPG key.

  • sudo cp /var/cudnn-local-repo-*/cudnn-local-*-keyring.gpg /usr/share/keyrings/

Refresh the repository metadata.

  • sudo apt-get update

Install the runtime library.

  • sudo apt-get install libcudnn8

Install the developer library.

  • sudo apt-get install libcudnn8-dev

Install the code samples and the cuDNN library documentation.

  • sudo apt-get install libcudnn8-samples

Verification

python3
>>> import tensorrt
>>>

Note: Might require sudo apt-get install nvidia-cuda-toolkit

Clone this wiki locally