How to improve CPU/GPU utilization in Gazebo simulation calculations

[cmeng-gao]

Low CPU/GPU utilization in Gazebo Harmonic when simulating multiple sphere robots

Environment

• OS: Ubuntu 22.04
• CPU: Intel 14900
• GPU: RTX 4060 Ti
• Gazebo Version: Harmonic

Description

When testing physics simulation in a blank SDF world with hundreds of sphere robots falling from air to ground, I observed:

• Very low Real-Time Factor (RTF) of 0.19
• Only one CPU core being utilized
• GPU remains completely idle

Expected Behavior

Gazebo should utilize multiple CPU cores and GPU resources effectively to improve simulation performance when handling large numbers of physics objects.

Current Behavior

Poor multi-core CPU utilization and no GPU usage results in severely degraded performance.

Steps to Reproduce

1. Create a blank SDF world
2. Spawn hundreds of sphere robot models
3. Simulate them falling from air to ground
4. Observe system resource usage (CPU/GPU) and RTF

Additional Information

• Are there any configuration options to improve multi-core utilization?
• Is GPU acceleration supported for physics calculations in Gazebo Harmonic?
• Any workarounds to improve performance for mass object simulations?

[arjo129]

As of now Gazebo does not yet support GPU based physics simulation. While sensors are rendered on the gpu, physics takes place on the CPU. It may be possible to improve this in the future, but its a really hard task that would require building a new gz-physics backend.

[cmeng-gao]

As of now Gazebo does not yet support GPU based physics simulation. While sensors are rendered on the gpu, physics takes place on the CPU. It may be possible to improve this in the future, but its a really hard task that would require building a new gz-physics backend.

Thanks for your reply, but during the test, I found that even if the calculation is on the CPU, the parallel computing of the multi-core CPU is not fully utilized.