RoadData platform requires GPU-enabled HW for the most of the recognition tasks. Currently, we support NVIDIA GPU with the driver >r525 and CUDA compute capability 5.2 or higher. You can check the list of CUDA-enabled GPU cards here: https://developer.nvidia.com/cuda-gpus
Due to high flexibility of the recognition platform, and a broad range of tasks it solves, the exact hardware requirements should be reviewed for each particular project.
Here are examples of system requirement definitions for the typical use cases, which can be used as a basic reference.
2-8 cameras with standard recognition module:
CPU: Intel Core i5 RAM: 8Gb GPU: NVIDIA RTX 3060 OS: Ubuntu 20.04
16-32 cameras with advanced recognition module.
CPU: Intel Core i9 RAM: 32Gb GPU: NVIDIA RTX 4070 OS: Ubuntu 20.04
Embedded System-on-Module (SoM)
RoadData can run on small low-power AI systems based on NVIDIA Jetson compute units.
The smallest unit in the Jetson family is Jetson Orin Nano.
An industrial computer based on the advanced platform NVIDIA Jetson Xavier NX
This type of HW is ideal, if the task requires centralized collection of structured data from multiple locations with few cameras at each location. Embedded compute units take the role of the Server, and a regular server or a cloud compute engine takes the role of the Manager.