NVIDIA Jetson TX2 NX Benchmarks¶

• NVIDIA Jetson TX2 NX Benchmarks
  • Miniature USB3 camera with 18 Mpix connected to TX2 NX
    • MRTech CamViewer demo application
    • Results for one MU181CR-ON camera
    • MRTech dragonfly application
    • Results for one MU181CR-ON color camera
    • Results for two MU181CR-ON color cameras running simultaneously

Fig.1. Jetson TX NX module

NVIDIA® Jetson™ TX2 NX is the next step in AI performance for embedded and edge products after the predecessor - Jetson TX2 module.

It delivers up to 2.5X the performance of Jetson Nano while sharing its form-factor and pin compatibility, as well as that of Jetson Xavier™ NX.
This compact SoC (system on a chip) packs hardware accelerators for the entire AI pipeline, where NVIDIA JetPack™ SDK provides the tools needed for the application.
Custom AI network development is simplified with pre-trained AI models from NVIDIA NGC™ and the NVIDIA Transfer Learning Toolkit.
The containerized deployments make updates flexible and seamless.

The XIMEA team decided to test various cameras connected to the module with NX dev kit and the results are listed below.

Miniature USB3 camera with 18 Mpix connected to TX2 NX¶

Hardware components:

• Jetson TX2 NX module
• NX developer kit - two full speed USB3 ports
• XIMEA MU181CR-ON camera: 18 MP, 4896 x 3680, Color model with USB3 interface
• GPU desktop as a receiving station

MRTech CamViewer demo application¶

CamViewer is a demo application designed to acquire images from USB3/PCIe cameras connected to a computer and display them on the screen.
The application is used for demonstrations and internal purposes of MRTech SK.

Processing pipeline:

• Free run acquisition
• auto exposure
• auto white balance
• demosaicing 5x5
• rendering on a screen (60 Hz)

Results for one MU181CR-ON camera¶

Resolution	Bit depth	Frame rate	Processing latency	G2G latency
Full frame 4896 x 3680	8 bit	21 Fps	440 ms	150 ms
Full frame 4896 x 3680	12 bit	14 Fps	62 ms	210 ms
2x2 binning 2448 x 1842	8 bit	59 Fps	16 ms	85 ms
2x2 binning 2448 x 1842	12 bit	56 Fps	25 ms	100 ms

MRTech dragonfly application¶

Dragonfly application was designed by MRTech SK to receive RAW images from one or more USB3/PCIe cameras and stream the processed images via the network to a remote computer.
The application can run on NVIDIA Jetson modules, embedded or desktop computers with NVIDIA GPU.
The application is currently being used in several projects in aerial, ground drones, on vehicles, as well as in medical devices.

Processing pipeline on Jetson:

• 8-bit acquisition
• auto exposure
• auto white balance
• demosaicing 5x5
• H.265 encoding
• RTSP streaming

Processing pipeline on the receiving station:

• receiving the RTSP stream
• H.265 decoding
• rendering on a screen (144 Hz)

Results for one MU181CR-ON color camera¶

Resolution	Bit depth	Frame rate	Jetson processing	Receiving side	Bitrate	G2G latency
Full HD ROI 1920 x 1080, 2.1MP	8 bit	98 Fps	12 ms	7 ms	9800 Mbps	50 ms
Full HD ROI 1920 x 1080, 2.1MP	16 bit	98 Fps	14 ms	7 ms	9800 Mbps	60 ms
4K (UHD) 3840 x 2160, 8.3MP	8 bit	45 Fps	34 ms	12 ms	9800 Mbps	90 ms
4K (UHD) 3840 x 2160, 8.3MP	8 bit	30 Fps	42 ms	12 ms	9800 Mbps	130 ms

Results for two MU181CR-ON color cameras running simultaneously¶

Resolution	Bit depth	Frame rate	Processing on Jetson	Bitrate	Processing on receiving side	G2G latency
Full HD ROI 1920 x 1080, 2.1MP	8 bit	90 Fps	15 ms	9800 Mbps	7 ms	60 ms
Full HD ROI 1920 x 1080, 2.1MP	16 bit	60 Fps	19 ms	9800 Mbps	7 ms	65 ms
4K (UHD) 3840 x 2160, 8.3MP	8 bit	22 Fps	50 ms	9800 Mbps	12 ms	125 ms

Note: The bandwidth on TX2 NX is limited in case two cameras are connected.