[Question] Isaac Lab simulation fails on Blackwell GPU (6000 Pro) – PhysX GPU pipeline fallback to CPU

[Fritingo]

Issue Description

After installing Isaac Lab successfully, the simulation does not run on GPU.
When running the following command:

./isaaclab.sh -p scripts/reinforcement_learning/rsl_rl/train.py --task=Isaac-Ant-v0 --headless

the process starts normally, but falls back to CPU due to a PhysX GPU solver failure.

System Information

• OS: Ubuntu 22.04
• GPU: NVIDIA Blackwell 6000 Pro
• Driver version: 575
• Python version: 3.11 (inside .venv)
• Installation method: Followed pip installation guide

Logs / Error Message

[INFO][AppLauncher]: Using device: cuda:0
...
2025-09-09T06:34:39Z [7,511ms] [Warning] [omni.physx.plugin] PhysX warning: GPU solver pipeline failed, switching to software, FILE /builds/omniverse/physics/physx/source/simulationcontroller/src/ScScene.cpp, LINE 781
2025-09-09T06:34:39Z [7,511ms] [Warning] [omni.physx.plugin] PhysX warning: GPU Bp pipeline failed, switching to software, FILE /builds/omniverse/physics/physx/source/simulationcontroller/src/ScScene.cpp, LINE 789

What I Tried

Verified that CUDA is installed and GPU is detected (nvidia-smi shows GPU working).
The device cuda:0 is correctly recognized by Isaac Lab.
However, PhysX reports GPU pipeline failure and falls back to software simulation.

Expected Behavior

Simulation should run using the GPU pipeline without falling back to CPU.

Questions

• Is PhysX GPU pipeline supported on Blackwell (6000 Pro) GPUs yet?
• Does Driver 575 support Isaac Lab and PhysX GPU features on Blackwell?
• Do I need a specific driver or CUDA version for compatibility?
• Are there workarounds to force GPU simulation instead of CPU fallback?

[PeterL-NV]

Hi, PhysX is supported on Blackwell GPUs, including the 6000 Pro you're using. It's recommended to use the latest production branch drivers. See Technical Requirements for recommended driver versions. I would recommend to try R570 driver to see if you notice the same issue.

[Jak253]

I have the exact same problem in the Docker container. I'm using an RTX 5070 Ti with driver version 570.153.02. I have also tried the 575 driver, but without success.
Have you managed to get it running yet, or are you still having the error?

[BITZXC]

I have the same problem. I'm using an RTX 5070 with driver version 580.82.07

[fcten]

According to this issue, the reason for this error may be that PhysX does not yet support the latest version of the driver.

Downgrading the NVIDIA driver version to 570.195.03 worked for me.

My environment is Ubuntu 24.04 on a physical machine with an RTX 4090.

[xiaoniute]

I have the exact same problem. My environment is Ubuntu 22.04 with an RTX 6000 Ada. NVIDIA driver version is 580.95.05.

[xiaoniute]

Downgrading the NVIDIA driver version to 570.195.03 worked for me too