Isaacsim/IsaacLab is getting stuck in DGX server in headless mode with --enable_cameras

Omniverse Isaac Sim
1

6.0.0
5.1.0
5.0.0
4.5.0
4.2.0
4.1.0
4.0.0
4.5.0
2023.1.1
2023.1.0-hotfix.1
Other (please specify):

Operating System

Ubuntu 24.04
Ubuntu 22.04
Ubuntu 20.04
Windows 11
Windows 10
Other (please specify):

GPU Information

  • Model: B200
  • Driver Version: 580.126.09

Isaacsim/lab is getting stuck in DGX spark server in headless mode during --enable_camera code

With other sensors the system is working perfectly. IsaacLab version 2.3.0

Steps to Reproduce

  1. simple examples like python scripts/reinforcement_learning/rl_games/train.py --task=Isaac-Cartpole-RGB-Camera-Direct-v0 --headless --enable_camerasis not working in headless mode

Error Messages (getting stuck)

2026-04-26T13:37:26Z [13,937ms] [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
2026-04-26T13:37:26Z [13,937ms] [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

Screenshots or Videos

image
image1504×136 20.3 KB

2

Hi @isouvickc,

Thanks for the report. A few clarifying questions first:

1. System clarification: You mentioned “DGX Spark” with “GPU: B200” – DGX Spark
uses a GB10 Grace Blackwell Superchip (not a discrete B200). Could you confirm
whether you’re on:

  • DGX B200 (data center system with 8x B200 GPUs), or
  • DGX Spark (desktop system with GB10 chip)?

This matters because multi-GPU systems (DGX B200) can have device selection issues
that affect PhysX.

2. Driver version: You’re running 580.126.09. Have you tried the specific
driver versions recommended in the
Isaac Sim 5.1.0 requirements or Omniverse Technical Requirements

While your 580.126.09 is the same R580 branch, sub-version differences can still
matter, especially on newer GPU architectures like Blackwell.

That said, here’s what we know about the root cause:

What’s happening:

  1. PhysX’s GPU solver/broadphase pipeline fails to initialize on the B200
    (the “GPU solver pipeline failed, switching to software” warnings)
  2. When PhysX falls back to CPU, the camera rendering pipeline (created by
    --enable_cameras) still expects GPU-backed physics data via Fabric
  3. This mismatch causes a deadlock – the render pipeline waits for physics
    data that never arrives through the GPU path

Why non-camera sensors work: They don’t depend on the GPU Fabric path for
physics state synchronization.

3

Thank you for the reply.

Yes, I’m using a B200 DGX server, not Spark. Pardon my mistake.

Regarding the driver selection, I’ll look into that with the server admin.
But this is a cumbersome process overall. No one wants to downgrade the driver version.

Do you know whether a patch update is coming soon?

4

No, the most up-to-date supported and recommended versions are documented at Technical Requirements — Omniverse Developer Guide.