NVIDIA GPU Acceleration

The Scout Agent supports rendering on NVIDIA hardware to enable high-speed, real-time license plate and vehicle recognition. If your camera captures vehicles moving over 60 miles per hour, NVIDIA GPU Acceleration makes it possible to deliver exceptional accuracy at high rates of speed.

Rekor maintains binaries for CUDA 11.0 using CuDNN 8 on 64-bit Ubuntu Linux 20.04. On Ubuntu 18.04 the software runs on CUDA 10.0 using CuDNN 7. The Scout software has been compiled for CUDA hardware versions 5.2 5.3 6.0 6.1 6.2 7.0 7.2 7.5 8.0. You can verify your CUDA version here: https://developer.nvidia.com/cuda-gpus

PLEASE NOTE: CUDA 11.4 (AND ABOVE) IS NOT SUPPORTED.

First, install the NVIDIA GPU drivers via the official NVIDIA repository:

wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2004/x86_64/cuda-ubuntu2004.pin
sudo mv cuda-ubuntu2004.pin /etc/apt/preferences.d/cuda-repository-pin-600
sudo apt-key adv --fetch-keys https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2004/x86_64/7fa2af80.pub
sudo add-apt-repository "deb https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2004/x86_64/ /"
sudo apt-get update
sudo apt-get -y install cuda-11-0

After installation, you will likely need to reboot. Make sure that the drivers are installed using the following command:

nvidia-smi

Next, install the Scout NVIDIA acceleration package. Select install_nvidia from the menu:

bash <(curl https://deb.openalpr.com/install)

The following properties will be set automatically in /etc/openalpr/openalpr.conf:

hardware_acceleration = 1
gpu_id = 0
gpu_batch_size = 5

The batch size controls how many images are simultaneously processed by the GPU. The default value is usually the best balance of performance and GPU memory usage.

Rekor also maintains special builds for Jetson TX-1/TX-2 hardware on Jetpack v3.1. To install, follow the standard Linux install instructions. The defaults are configured such that the GPU will automatically be installed, enabled, and configured with a default batch size of 5.

Last updated