Hello everyone,
I have recently encountered issues related to GPU usage and performance degradation while developing a ROS2 project using NVIDIA Isaac Sim 4.0 on Ubuntu 22.04 with an NVIDIA RTX 3090 graphics card. Isaac Sim is installed on a 2TB hard drive. Below are the details of my issues and a specific question regarding the Visual SLAM tutorial:
- Tutorials and Actions:
- I followed the “URDF Import: Turtlebot” tutorial to successfully import the URDF model of the robot along with a simple scene.
- Following the “Driving TurtleBot via ROS2 messages” and “ROS2 Cameras” tutorials, I added the publication of
/cmd_veland/rgbtopics through Omnigraph. - I also tried running the scenario as per the Visual SLAM tutorial. After loading the scene and hitting play, the system became very laggy with GPU utilization reaching 100%.
- Issue Description:
- When running the
/cmd_veltopic alone, the GPU utilization is at 62%. - However, adding the
/rgband/depthcamera topics caused the GPU utilization to surge to 96%, resulting in noticeable system lag. - The situation further deteriorated when attempting the Visual SLAM tutorial, where GPU utilization maxed out at 100% upon starting the simulation.
- Questions:
- Has anyone experienced similar issues where just adding a camera’s
/rgband/depthtopics leads to significant performance degradation? - Does running the Visual SLAM scenario typically result in such high resource usage, and how can this be mitigated?
- Are there any recommendations or tips for optimizing or maintaining smooth operation of the simulation environment when additional topics or computational demands are introduced?
I am looking for any possible optimization advice or configuration adjustments to address these performance issues. Any insights or suggestions would be greatly appreciated!