ubuntu16.04 whole process of cuda11.3, cudnn8.2.1 and tensorrt8.0.3.4 installation

Before installation, please explain a few points. Because you are not familiar with the installation method of linux, you can see that there are many installation methods, including source code compilation, dpkg, apt, etc. we all know the source code method here, compile through Makefile or cmake, and then install. The process is complex and error prone. Apt is used in conjunction with dpkg, Here is a brief introduction dpkg:

dpkg

linux developers compile the software to be installed on the fixed hardware platform and operating system platform, and then package all relevant files of the software into a special format file. This file also includes the pre detection system and dependent scripts, and provides all file information provided by the software. After the customer obtains the software package, as long as it is installed through a specific command, the software will detect the existence of the precursor software according to the internal script. If the installation environment meets the requirements, the installation will start. After installation, the information of the software will be written into the software management mechanism to complete future upgrades, deletions and other operations.

Here are dpkg some basic commands:

  1. dpkg -i package-name   Install package
  2. dpkg -r package-name   Delete package (keep configuration information)
  3. dpkg -P package-name   Delete package (including configuration information)
  4. dpkg --configure package-name   Configure software packages. If - a is added, all unconfigured software packages are configured
  5. dpkg --unpack package-name   Unpack the software package to the system directory, but do not configure it

The following commands are used to query the software package

  1. dpkg -I filename   View the software description (use dpkg -l directly)   Query all installed packages. The filename can be regular. I usually use ` ` ` dpkg -l | grep "filename" because the software name will not be remembered)
  2. dpkg -L filename   View the package installation files and directories corresponding to package name
  3. dpkg -s filename   View the package information corresponding to package name
  4. dpkg -S filename-pattern   Find the package name containing filename from the installed package

Introduction to dpkg software package related documents
/etc/dpkg/dpkg.cfg   Configuration file of dpkg package management software
/var/log/dpkg.log   Log file of dpkg package management software
/var/lib/dpkg/available   Store all installed software package information of the system
/var/lib/dpkg/status   Store the status information of all installed software in the system
/var/lib/dpkg/info   Record the control information file of the installation package control directory

apt-get

Although we have solved a lot of problems in the software installation process when using dpkg, we still need to solve them manually when the dependencies are not satisfied. apt solves such problems. linux distribution first places the software in the corresponding server, then analyzes the software dependencies and records them, Then, when the client needs to install the software, compare the list with the existing software data of the local dpkg, and you can get all the required software with dependent attributes from the network.

Here are some basic commands of apt:

  1. apt-get update   Update source
  2. apt-ge dist-upgrade   Upgrade system
  3. apt-get upgrade   Update all installed packages
  4. apt-get install package_name   Install the package (plus -- reinstall)
  5. apt-get remove   Remove package (retain configuration information)
  6. apt-get purge package_name   Remove package (delete configuration information)
  7. apt show pack_name   Get information about the package
  8. apt search page_name   Search for information about packages
  9. apt-cache depends package   Understanding usage dependencies
  10. apt-get check   Check for damaged dependencies

Introduction to apt package related documents:
/etc/apt/sources.list   Record the address of the software source
/var/cache/apt/archives   The downloaded packages are placed here

The above are several installation methods of linux. What does it have to do with our installation of NVIDIA series? In fact, NVIDIA provides many installation methods. Each person adopts different methods, which makes people who don't know linux very prone to errors. Therefore, let's introduce it first and start installing this library

Installation by dpkg

cuda: 11.3.1

wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu1604/x86_64/cuda-ubuntu1604.pin
sudo mv cuda-ubuntu1604.pin /etc/apt/preferences.d/cuda-repository-pin-600
wget https://developer.download.nvidia.com/compute/cuda/11.3.1/local_installers/cuda-repo-ubuntu1604-11-3-local_11.3.1-465.19.01-1_amd64.deb
sudo dpkg -i cuda-repo-ubuntu1604-11-3-local_11.3.1-465.19.01-1_amd64.deb
sudo apt-key add /var/cuda-repo-ubuntu1604-11-3-local/7fa2af80.pub
sudo apt-get update
sudo apt-get -y install cuda

cudnn: 8.2.1

  You need to download these four files, otherwise you will not be able to install successfully

  1. Install runtime libraries, for example: 
    sudo dpkg -i libcudnn8_x.x.x-1+cudax.x_amd64.deb

    perhaps

    sudo dpkg -i libcudnn8_x.x.x-1+cudax.x_arm64.deb
  2. Install the developer library, for example: 
    sudo dpkg -i libcudnn8-dev_8.x.x.x-1+cudax.x_amd64.deb

    perhaps

    sudo dpkg -i libcudnn8-dev_8.x.x.x-1+cudax.x_arm64.deb
  3. Install code samples and cuDNN library documentation, for example: 
    sudo dpkg -i libcudnn8-samples_8.x.x.x-1+cudax.x_amd64.deb

    perhaps

    sudo dpkg -i libcudnn8-samples_8.x.x.x-1+cudax.x_arm64.deb

    4. After I installed the above, my own test failed, so I need to install it   cuDNN Library for Linux (x86_64) , which is the first one in the figure above:

    Unzip the cuDNN package.

     tar -xzvf cudnn-x.x-linux-x64-v8.x.x.x.tgz

    perhaps

     tar -xzvf cudnn-x.x-linux-aarch64sbsa-v8.x.x.x.tgz
    Copy the following files to the CUDA Toolkit directory. 
    sudo cp cuda/include/cudnn*.h /usr/local/cuda/include 
sudo cp -P cuda/lib64/libcudnn* /usr/local/cuda/lib64 
sudo chmod a+r /usr/local/cuda/include/cudnn*.h /usr/local/cuda/lib64/libcudnn*

tensorrt: 8.0.3

  1. Install TensorRT from the local. dep package. replace   ubuntuxx04,   cudax.x,   trt8.xxx   and   yyyymmdd   Use your specific operating system version, CUDA version, TensorRT version, and package date. 
    os="ubuntuxx04"
tag="cudax.x-trt8.x.x.x-yyyymmdd"
sudo dpkg -i nv-tensorrt-repo-${os}-${tag}_1-1_amd64.deb
sudo apt-key add /var/nv-tensorrt-repo-${os}-${tag}/7fa2af80.pub

sudo apt-get update
sudo apt-get install tensorrt
    If you use Python 3.x: 
    sudo apt-get install python3-libnvinfer-dev
    The following additional packages will be installed: 
    python3-libnvinfer
    If you plan to use TensorRT with TensorFlow: 
    sudo apt-get install uff-converter-tf

    this   graphsurgeon-tf   The software package will also be installed using the above command.

    If you want to run an example that requires ONNX   Figure surgeon   Or use the Python module in your own project to run: 
    sudo apt-get install onnx-graphsurgeon
  2. Verify installation. 
    dpkg -l | grep TensorRT
    You should see something similar to the following: 
    ii  graphsurgeon-tf	8.2.0-1+cuda11.4	amd64	GraphSurgeon for TensorRT package
ii  libnvinfer-bin		8.2.0-1+cuda11.4	amd64	TensorRT binaries
ii  libnvinfer-dev		8.2.0-1+cuda11.4	amd64	TensorRT development libraries and headers
ii  libnvinfer-doc		8.2.0-1+cuda11.4	all	TensorRT documentation
ii  libnvinfer-plugin-dev	8.2.0-1+cuda11.4	amd64	TensorRT plugin libraries
ii  libnvinfer-plugin8	8.2.0-1+cuda11.4	amd64	TensorRT plugin libraries
ii  libnvinfer-samples	8.2.0-1+cuda11.4	all	TensorRT samples
ii  libnvinfer8		8.2.0-1+cuda11.4	amd64	TensorRT runtime libraries
ii  libnvonnxparsers-dev		8.2.0-1+cuda11.4	amd64	TensorRT ONNX libraries
ii  libnvonnxparsers8	8.2.0-1+cuda11.4	amd64	TensorRT ONNX libraries
ii  libnvparsers-dev	8.2.0-1+cuda11.4	amd64	TensorRT parsers libraries
ii  libnvparsers8	8.2.0-1+cuda11.4	amd64	TensorRT parsers libraries
ii  python3-libnvinfer	8.2.0-1+cuda11.4	amd64	Python 3 bindings for TensorRT
ii  python3-libnvinfer-dev	8.2.0-1+cuda11.4	amd64	Python 3 development package for TensorRT
ii  tensorrt		8.2.0.x-1+cuda11.4 	amd64	Meta package of TensorRT
ii  uff-converter-tf	8.2.0-1+cuda11.4	amd64	UFF converter for TensorRT package
ii  onnx-graphsurgeon   8.2.0-1+cuda11.4  amd64 ONNX GraphSurgeon for TensorRT package

adopt   dpkg installation is very convenient, but this method is not recommended because you can't determine where you installed it and it's inconvenient to find it later. Therefore, the following installation is recommended.

runfile installation

cuda11.3.1 

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

cudnn: 8.2.1

Download the first one directly:   Installation required   cuDNN Library for Linux (x86_64) , which is the first one in the figure above:

  Unzip the cuDNN package.

 tar -xzvf cudnn-x.x-linux-x64-v8.x.x.x.tgz

perhaps

 tar -xzvf cudnn-x.x-linux-aarch64sbsa-v8.x.x.x.tgz

Copy the following files to the CUDA Toolkit directory.

sudo cp cuda/include/cudnn*.h /usr/local/cuda/include 
sudo cp -P cuda/lib64/libcudnn* /usr/local/cuda/lib64 
sudo chmod a+r /usr/local/cuda/include/cudnn*.h /usr/local/cuda/lib64/libcudnn*

tensorrt: 8.0.3

 

  1. Unzip the tar file. 
    version="8.x.x.x"
arch=$(uname -m)
cuda="cuda-x.x"
cudnn="cudnn8.x"
tar xzvf TensorRT-${version}.Linux.${arch}-gnu.${cuda}.${cudnn}.tar.gz
    where:
    • 8.xxx is your TensorRT version
    • cuda-xx   CUDA version   ten point two   perhaps   eleven point four
    • cudnn8.x   Is cuDNN version   eight point two
    The directory will have subdirectories, such as   Library,   include,   Data, wait 
    ls TensorRT-${version}
bin  data  doc  graphsurgeon  include  lib  onnx_graphsurgeon  python  samples  targets  TensorRT-Release-Notes.pdf  uff
  2. Add absolute path library for TensorRT   Directory to environment variable   LD_LIBRARY_PATH: 
    export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:<TensorRT-${version}/lib>
  3. Install Python   TensorRT round file. 
    cd TensorRT-${version}/python

python3 -m pip install tensorrt-*-cp3x-none-linux_x86_64.whl
  4. Install Python UFF round file. This is only necessary if you plan to use TensorRT with TensorFlow. 
    cd TensorRT-${version}/uff

python3 -m pip install uff-0.6.9-py2.py3-none-any.whl
    Check installation: 
    which convert-to-uff
  5. Install Python   Call. 
    cd TensorRT-${version}/graphsurgeon

python3 -m pip install graphsurgeon-0.4.5-py2.py3-none-any.whl
  6. Install Python   onnx-graphsurgeon   Round file. 
    cd TensorRT-${version}/onnx_graphsurgeon
	
python3 -m pip install onnx_graphsurgeon-0.3.12-py2.py3-none-any.whl

        

Install CUDA 11.0, cuDNN 8.0.5, PyTorch 1.7.0 for Ubuntu 20.04

NVIDIA cuDNN v8 deb method installation and uninstallation tutorial (Linux/Ubuntu)

Ubuntu 1804 installation of cuda, cudnn and TensorRT

Keywords: Linux Operation & Maintenance CUDA

Added by Fantast on Tue, 07 Dec 2021 20:56:40 +0200

Popular Keywords

©2024 Programming VIP