System Configuration
- Primary Hardware: Jetson Orin AGX (64GB) - experiencing issues
- Comparison Hardware: Jetson Orin NX on ARK Jetson PAB Carrier ( ARK Jetson PAB Carrier | ARK Documentation ) - performing better
- Storage: NVMe SSD with JetPack 6.2
- ROS: ROS2 Humble
- Isaac ROS: Isaac ROS Visual SLAM package
- Navigation: Nav2 with custom polygon grid coverage planner
Problem Description
I’m experiencing counterintuitive performance issues where a Jetson Orin NX significantly outperforms a Jetson Orin AGX running the same Isaac ROS + Nav2 stack. This contradicts expected performance hierarchy.
Performance Issues on Orin AGX:
- First execution: Acceptable performance
- Subsequent runs: Severe performance degradation (2nd/3rd navigation attempts)
- Overall speed: Much slower than Orin NX running identical software stack
Network Performance Disparity:
Orin AGX speedtest results:
Download: 17.59 Mbit/s
Upload: 1.70 Mbit/s
Laptop on same network: ~100 Mbit/s up/down
Software Stack (Identical on Both Systems)
- Isaac ROS Visual SLAM with ZED 2i stereo camera
- Nav2 navigation stack
- RPLiDAR S2 integration
- Custom polygon grid planner for autonomous coverage
- Key difference : The Ark Jetson Carrier PAB uses a PX4 based low level controller to control the motors. The Jetson Orin AGX uses a custom ROS2 motor driver to get encoder data and ensure each wheel is moving at the required rpm.
Key Questions
- Hardware Performance Inversion: Why would an Orin NX outperform an Orin AGX on the same Isaac ROS workload?
- Network Bottleneck: Could the slow network performance on Orin AGX be affecting ROS2 DDS communication and overall system performance?
- JetPack 6.2 on AGX: Are there known performance regressions or configuration issues with JetPack 6.2 on Orin AGX specifically?
Configuration Details
- Using Isaac ROS development container
- Both systems running identical launch configurations
The fact that the supposedly less powerful Orin NX consistently outperforms the AGX suggests there may be a configuration, driver, or firmware issue specific to the AGX platform.
Any insights on debugging this performance inversion would be greatly appreciated! Would be more than happy to provide further details.