The recommended hardware for Stream is as follows:
ARCHITECTURE. Stream runs on a variety of devices that support ARMv7, ARMv8 and x86 machines. Some examples below:
- ARMv7, ARMv8: Nvidia Jetson Nano, Nvidia Jetson Xavier NX, Raspberry Pi 4, Banana Pi M4, Orange Pi, RockPro64, Khadas VIM3, Odroid N2, Odroid XU4, ASUS Tinker Board S, ASUS Tinker Board.
- x86: LattePanda Alpha, LattePanda Delta, UDOO Bolt V8, UP Squared, ASUS PN40.
CPU. The main drivers of CPU utilization is the amount of vehicles observed by Stream. No GPU is needed. See examples of machines with a certain Passmark score.
- For High Traffic (e.g. highway, busy street): CPU Passmark Score = 1200 * # Stream Cameras.
- For Low Traffic (e.g. parking entrance, small street): CPU Passmark Score = 700 * # Stream Cameras.
To figure out the Passmark of an existing machine, just do an online search! This Google Search: “intel i5-8250u cpu @ 1.60ghz passmark” returns Passmark 5942.
RAM. The main drivers of RAM utilization is the video size and Vehicle Make Model Color (and Orientation) identification.
- 1920×1080 video with MMC: RAM = 450MB * cameras.
- 1920×1080 video without MMC: RAM = 400MB * cameras.
- 1280 x 720 video with MMC: RAM = 400MB * cameras
- 1280×720 video without MMC: RAM = 350MB * cameras.
Hard Drive Space. Stream has a fairly small footprint. The main occupants of the hard drive will be the jpg images captured and CSV from the ALPR. If you have 0.5 GB of storage, that would be plenty.
eFPS. You can set the eFPS based on the speed of the vehicles to optimize system resources.
- For vehicles traveling <30 mph → set the eFPS to 4-5.
- For vehicles traveling 30-60 mph → set eFPS to 8-10.
- For vehicles traveling >60 mph → set eFPS to 12.
- To compute eFPS, divide the camera FPS by the Sample Rate that you set in Stream. So for example, if the camera has 30 FPS, then set sample=6 to get an eFPS = 5.
EXAMPLES
- Nvidia Jetson Xavier AGX with GPU can support 3 Stream cameras with MMC.
- For 2 Stream camera with MMC deployment for High Traffic, an Intel Core i3-4100M @ 2.5 GHz (Passmark 2440) with 1 GB RAM should be sufficient.
- For a 20 Stream camera deployment for highway monitoring with Vehicle MMC, you will want a machine with 24000 Passmark Score, such as an Intel Core i7-11700K @ 3.60GHz.
- Raspberry Pi 4 with 4 GB RAM can support 1 Stream camera with MMC at 1080p in High Traffic.
- Raspberry Pi 5 with 4 GB RAM can support 2 Stream cameras without MMC at 720p in Low Traffic.
- Jetson Nano with 4 GB RAM can support 1 Stream camera. Nano with 8 GB can support 2-3 Stream cameras.
- See more examples here!
