-
Notifications
You must be signed in to change notification settings - Fork 368
Description
Description
We are having issues with Isaac sim performance when the simulation is run, FPS degrades as we add more and more iw.hubs. We have 3 iw.hubs in the scene and we are down to ~10 FPS, and when we add iw.hub without LiDARS and Cameras it doesn't degrade the performance. By adding each sensor one by one performance slowly falls of. Things we noticed and tried:
- https://forums.developer.nvidia.com/t/cpu-bottleneck/321542
- https://forums.developer.nvidia.com/t/isaac-sim-poor-cpu-threads-utilization/312338
- https://forums.developer.nvidia.com/t/isaac-sim-performance/277249
- https://forums.developer.nvidia.com/t/the-problems-relevant-to-the-simulation-time-step-and-how-to-speed-up/302729
- [Question] Low GPU Utilization vs High Single-Core CPU Usage During RL Training IsaacLab#3043
-https://www.reddit.com/r/reinforcementlearning/comments/1npzkp2/isaaclab_sims_getting_bottlenecked_by_singlecore/
Btop/Htop when the simulation is running:
Should this look like this or is this the bottleneck? We've tried running the scene that has the 60 FPS and it looks about the same
We've tried adding these options into the config
--/persistent/physics/numThreads=32
--/plugins/omni.tbb.globalcontrol/maxThreadCount=32
--/plugins/carb.tasking.plugin/threadCount=32
We've tried using the 'Scene optimizer' extension, unfortunately to no avail. These are statistics of our scene:
The only thing where we noticed a differnce is when using 'Fabric,' although that only increases FPS slighlty (by 5-6) which ofcourse would be futile when adding more and more robots.
To replicate this, only adding 3-4 iw.hubs from Create > ROS 2 Assets > iw.hub ROS into the empty stage and play the simulation.
Isaac Sim version
5.1.0
Operating System (OS)
Ubuntu 24.04
GPU Name
RTX PRO 6000 Blackwell Max-Q Workstation Edition
GPU Driver and CUDA versions
580.126.09, CUDA Version : 13.0
Logs
No response
Additional information
No response