Build OpenCV from source
Install Math Kernel Library (MKL/BLAS/LAPACK/OPENBLAS)
You are recommended to install all Math Kernel Library and then compile framework (e.g pytorch, mxnet) from source using custom config for optimization.Install all LAPACK+BLAS:
sudo apt install libjpeg-dev libpng-dev libblas-dev libopenblas-dev libatlas-base-dev liblapack-dev liblapacke-dev gfortran
Install MKL:
# Get the key
wget https://apt.repos.intel.com/intel-gpg-keys/GPG-PUB-KEY-INTEL-SW-PRODUCTS-2019.PUB
# now install that key
apt-key add GPG-PUB-KEY-INTEL-SW-PRODUCTS-2019.PUB
# now remove the public key file exit the root shell
rm GPG-PUB-KEY-INTEL-SW-PRODUCTS-2019.PUB
# Add to apt
sudo wget https://apt.repos.intel.com/setup/intelproducts.list -O /etc/apt/sources.list.d/intelproducts.list
sudo sh -c 'echo deb https://apt.repos.intel.com/mkl all main > /etc/apt/sources.list.d/intel-mkl.list'
# Install
sudo apt-get update
sudo apt-get install intel-mkl-2020.4-912
Fresh install NVIDIA driver (PC/Laptop/Workstation)
# Remove old packages
sudo apt-get remove --purge '^nvidia-.*'
sudo apt-get install ubuntu-desktop
sudo apt-get --purge remove "*cublas*" "cuda*"
sudo apt-get --purge remove "*nvidia*"
sudo add-apt-repository --remove ppa:graphics-drivers/ppa
sudo rm /etc/X11/xorg.conf
sudo apt autoremove
sudo reboot
# After restart
sudo ubuntu-drivers devices
sudo ubuntu-drivers autoinstall
sudo reboot
NVIDIA-SMI has failed because it couldn’t communicate with the NVIDIA driver
First, make sure that you have "Fresh install NVIDIA driver". If not work, try this bellow
- Make sure the package nvidia-prime is installed:
sudo apt install nvidia-prime
Afterwards, run
sudo prime-select nvidia
- Make sure that NVIDIA is not in blacklist
grep nvidia /etc/modprobe.d/* /lib/modprobe.d/*
to find a file containing blacklist nvidia and remove it, then run
sudo update-initramfs -u
- If get error
This PCI I/O region assigned to your NVIDIA device is invalid:
sudo nano /etc/default/grub
edit GRUB_CMDLINE_LINUX_DEFAULT="quiet splash pci=realloc=off"
sudo update-grub
sudo reboot
Check current CUDA version
nvcc --version
Check current supported CUDA versions
ls /usr/local/
Select GPU devices
CUDA_VISIBLE_DEVICES=<index-of-devices> <command>
CUDA_VISIBLE_DEVICES=0 python abc.py
CUDA_VISIBLE_DEVICES=0 ./sample.sh
CUDA_VISIBLE_DEVICES=0,1,2,3 python abc.py
CUDA_VISIBLE_DEVICES=0,1,2,3 ./sample.sh
Switch CUDA version
CUDA_VER=11.3
export PATH="/usr/local/cuda-$CUDA_VER/bin:$PATH"
export LD_LIBRARY_PATH=/usr/local/cuda-$CUDA_VER/lib64${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}
Check NVENV/NVDEC status
nvidia-smi dmon
see the tab %enc and %dec
Error with distributed training NCCL (got freezed)
export NCCL_P2P_DISABLE="1"
Install CMake from source
version=3.23
build=2 ## don't modify from here
mkdir ~/temp
cd ~/temp
wget https://cmake.org/files/v$version/cmake-$version.$build.tar.gz
tar -xzvf cmake-$version.$build.tar.gz
cd cmake-$version.$build/
./bootstrap
make -j8
sudo make install
Install MXNet from source (for AMD CPU & NVIDIA GPU)
git clone --recursive --branch 1.9.1 https://github.com/apache/incubator-mxnet.git mxnet
cd mxnet
cp config/linux_gpu.cmake config.cmake
rm -rf build
mkdir -p build && cd build
cmake -DUSE_CUDA=ON -DUSE_CUDNN=OFF -DUSE_MKL_IF_AVAILABLE=OFF -DUSE_MKLDNN=OFF -DUSE_OPENMP=OFF -DUSE_OPENCV=ON -DUSE_BLAS=open ..
make -j32
cd ../python
pip install --user -e .
Tensorflow could not load dynamic library 'cudart64_101.dll'
For above example tensorflow would require CUDA 10.1, please switch to CUDA 10.1 or change tensorflow version which compatible with CUDA version, check here: https://www.tensorflow.org/install/source#gpuFix Deepstream (6.2+) FFMPEG OpenCV installation
Fix some errors about undefined reference & not found of libavcodec, libavutil, libvpx, ...apt-get install --reinstall --no-install-recommends -y libavcodec58 libavcodec-dev libavformat58 libavformat-dev libavutil56 libavutil-dev gstreamer1.0-libav
apt install --reinstall gstreamer1.0-plugins-good
apt install --reinstall libvpx6 libx264-155 libx265-179 libmpg123-0 libmpeg2-4 libmpeg2encpp-2.1-0
gst-inspect-1.0 | grep 264
rm ~/.cache/gstreamer-1.0/registry.x86_64.bin
apt install --reinstall libx264-155
apt-get install gstreamer1.0-libav
apt-get install --reinstall gstreamer1.0-plugins-ugly
Gstreamer pipeline to convert MP4-MP4 with re-encoding
gst-launch-1.0 filesrc location="<path-to-input>" ! qtdemux ! video/x-h264 ! h264parse ! avdec_h264 ! videoconvert ! x264enc ! h264parse ! qtmux ! filesink location=<path-to-output>
Gstreamer pipeline to convert RTSP-RTMP
gst-launch-1.0 rtspsrc location='rtsp://<path-to-rtsp-input>' ! rtph264depay ! h264parse ! flvmux ! rtmpsink location='rtmp://rtmp://<path-to-rtmp-output>'
Gstreamer pipeline to convert RTSP-RTMP with reducing resolution
gst-launch-1.0 rtspsrc location='rtsp://<path-to-rtsp-input>' ! rtpbin ! rtph264depay ! h264parse ! avdec_h264 ! videoconvert ! videoscale ! video/x-raw,width=640,height=640 ! x264enc ! h264parse ! flvmux streamable=true ! rtmpsink location='rtmp://<path-to-rtmp-output>'