Jetson Orin Nano AI Performance Comparison and Camera Support

Issue Overview

Users have raised questions regarding the AI performance comparison between the Jetson Orin Nano and other Jetson models, specifically the TX2NX. They are seeking clarification on the performance metrics used for the Orin Nano, as it is not expressed in TFLOPs like the TX2NX. Additionally, there are inquiries about the video encoding capabilities and camera input support of the Orin Nano.

Possible Causes

The differences in performance metrics between the Jetson models can be attributed to the following factors:

1. Hardware Differences: The Jetson TX1, TX2, and Nano support FP16/FP32 hardware, which is measured in TFLOPS. On the other hand, the Jetson Xavier and Orin models support INT8 hardware in addition to FP16/FP32, and their performance is measured in TOPS.

2. Architectural Enhancements: The Jetson Orin Nano features both Tensor Cores and GPU Cores, with Tensor Cores being measured in TOPS. This architectural difference contributes to the varying performance metrics used for the Orin Nano compared to older Jetson models.

Troubleshooting Steps, Solutions & Fixes

To address the concerns regarding AI performance comparison and camera support for the Jetson Orin Nano, consider the following steps:

1. Emulate Jetson Orin Nano Configuration: As FP16 benchmarks for the Orin Nano have not been officially published, it is recommended to use the Jetson AGX Orin Developer Kit to emulate the Orin Nano configuration. This will allow you to run the necessary benchmarks and assess the FP16 performance of the Orin Nano.

2. Refer to the Jetson Orin Nano Datasheet: To determine the supported camera input formats for the Orin Nano, consult the MIPI CSI Camera Interface section of the Orin Nano datasheet. The datasheet lists the supported input data formats, which include:

   • RGB: RGB888, RGB666, RGB565, RGB555, RGB444
   • YUV: YUV420-8b (legacy), YUV420-8b
   • RAW: RAW6, RAW7, RAW8, RAW10, RAW12, RAW14, RAW16

   Note that 20-bit Bayer camera input is not explicitly listed in the datasheet. If further clarification is needed, it is recommended to start a new topic specifically addressing camera support to receive assistance from camera experts.

3. Evaluate Video Encoding Performance: The video encoding performance of the Orin Nano depends on the library and settings used, such as the x264enc GStreamer element. To assess the video encoding capabilities, experiment with different libraries and configurations to determine the supported resolutions and codecs.

4. Consult Additional Resources: Refer to the Jetson Benchmarks page on the NVIDIA Developer website for more information on the performance of various Jetson models. While a direct comparison between the TX2NX and Orin Nano may not be available, the benchmarks provide insights into the capabilities of different Jetson modules.

By following these steps and leveraging the available resources, you can gain a better understanding of the AI performance, camera support, and video encoding capabilities of the Jetson Orin Nano. If further assistance is required, consider reaching out to the NVIDIA developer community or starting new topics focused on specific aspects of the Orin Nano’s functionality.