[Bug Report] RTX 5090: Blurry, Pixelated and Noisy Rendering in RTX Real-Time Mode on Ubuntu 22.04 (Driver 570.133.07 / 570.124.04)

[renhong-zhang]

Describe the bug

GPU: 5090(This bug relates to 50-series card)
When using RTX - Real Time rendering mode in Isaac Sim on Ubuntu 22.04, the rendered robot appears extremely blurry, pixelated, and noisy. The issue presents as:

• The robot body has a fuzzy, voxel-like appearance
• The shadow is heavily pixelated and inaccurate

This occurs with both default models and USD files I import. I have not modified any materials or shaders. Switching to Windows 11 (driver 572.83) on the same machine renders the scene correctly, indicating this issue is specific to Ubuntu + 50-series GPU + RTX Real-Time mode.

Steps to reproduce

1. Launch Isaac Sim 4.5.0 on Ubuntu 22.04
2. Load any default robot scene or asset (or import a USD model)
3. Switch to RTX - Real Time mode
4. Observe pixelated rendering of robot and shadows

Mode: RTX - Real-Time

Mode: RTX - Real-Time 2.0

Mode: Interactive(Path Tracing)

System Info

• Isaac Sim Version: 4.5.0
• OS: Ubuntu 22.04
• GPU: RTX 5090
• CUDA: 12.8
• GPU Driver: 570.133.07 and 570.124.04

Additional context

• The same issue does not occur on Windows 11 (driver 572.83)
• Please prioritize a fix or provide a workaround — thank you!

Checklist

• I have checked that there is no similar issue in the repo (required)
• I have confirmed that the issue is not related to custom shaders or materials

Acceptance Criteria

• Robots and shadows are rendered clearly in RTX - Real Time mode on Ubuntu 22.04
• Rendering matches Windows behavior using the same scene and model

[RandomOakForest]

Thanks for posting this. The team will investigate.

[Nate711]

Same here

[ShuoYangRobotics]

same here.

[isso313]

Same

[jwjoel]

Same here

[Nate711]

I added the following cli args to rsl_rl/train.py and play.py so that path tracing is automatically used with low samples per frame.

With 1 sample per pixel per frame it's just as fast as RTX-Real-Time for me.

For those wondering why this works, these cli_args are passed to AppLauncher, which in turn passes the args to SimulationApp which supports these parameters https://docs.omniverse.nvidia.com/py/isaacsim/source/extensions/omni.isaac.kit/docs/index.html

parser.add_argument(
    "--renderer",
    type=str,
    default="PathTracing",
    choices=["RayTracedLighting", "PathTracing"],
    help="Renderer to use.",
)
parser.add_argument(
    "--samples_per_pixel_per_frame",
    type=int,
    default=1,
    help="Number of samples per pixel per frame.",
)

[renhong-zhang]

I'm currently using RTX - Interactive mode as a workaround for simpler scenes, but it has poor visual quality and performance in indoor scenarios. According to a recent NVIDIA Developer Forum update, the issue with RTX - Real-Time rendering stems from compatibility problems between NVIDIA Blackwell GPUs and the Kit version in Isaac Sim (4.5.0 and 4.2.0). As Isaac Sim is closed-source, we have to wait for an official update to address this.

[zhaowy312]

@renhong-zhang My system configuration is the same as yours, and I recently tried to use isaac lab.

In the official documentation, they mentioned For 50 series GPUs,please use the latest PyTorch nightly build instead of PyTorch 2.5.1,which comes with Isaac Sim: pip install--upgrade--pre torch torchvision--index-url https://download.pytorch.org/whl/nightly/cu128.

. After executing./isaaclab.sh--install rl_games, I followed their instructions, but encountered version conflict:isaaclab 0.36.6 requires torch==2.5.1,but you have torch 2.8.0.dev20250414+cu128 which is incompatible.

isaaclab-rl 0.1.4 requires torch==2.5.1, but you have torch 2.8.0.dev20250414+cu128 which is incompatible.

isaaclab-tasks 0.10.27 requires torch==2.5.1, but you have torch 2.8.0.dev20250414+cu128 which is incompatible.

I don't know if you have encountered this problem, please give me some help, thank you very much.

[Nate711]

You can remove the version specification from the setup.py files in isaaclab-rl and isaaclab-tasks.

[zhaowy312]

@Nate711 It works！Many thanks！

[singh-jayant]

Weird, but pleasantly surprising observations

Observation 1

When I run IsaacSim with isaac-sim.sh or isaac-sim.selector.sh, I get blurry rendering in the viewport with Real-Time and Real-Time 2.0 rendering. Auto option on DLSS works better, so does DLAA, but all still noisy. Interactive mode works best, but incredibly slow.

Observation 2

Replicator data generation also generates noisy RGB images, but depth images are perfectly clean

Observation 3

BUT, when I use scripts/tutorials/00_sim/set_rendering_mode.py from IsaacLab to launch the exact same USD, the rendering works clean and fast, no issues at all. Not sure why or how but I ain't complaining.

[atyagi702]

I am using the exact same setup as described by OP but on Ubuntu 20.04 and it seems to get rid of the noise. But I am still facing the issue in Ubuntu 22.04.

[snknitheesh]

I am using the same setup with isaac-4.5. Facing the same issue. I even installed the apt version required for torch and torchvision.

Any help would be very good. I get the same output in interactive and real-time modes.

[Nate711]

@singh-jayant @snknitheesh @atyagi702 Read the above message by @renhong-zhang
I'm currently using RTX - Interactive mode as a workaround for simpler scenes, but it has poor visual quality and performance in indoor scenarios. According to a recent [NVIDIA Developer Forum update](https://forums.developer.nvidia.com/t/the-isaac-sim-gui-is-blurry/327759/14), the issue with RTX - Real-Time rendering stems from compatibility problems between NVIDIA Blackwell GPUs and the Kit version in Isaac Sim (4.5.0 and 4.2.0). As Isaac Sim is closed-source, we have to wait for an official update to address this.

[xuxuezhou]

I am facing the following error when I run scripts without headless. I have install the latest PyTorch nightly build. Wondering if anyone have the same issue and know how to solve it, thanks!

RuntimeError: nvrtc: error: invalid value for --gpu-architecture (-arch)

System info

• Isaac Sim Version: 4.5.0
• OS: Ubuntu 22.04
• GPU: RTX 5090
• CUDA: 12.8
• GPU Driver: 570.153.02

[Hahalim2022y]

I am facing the following error when I run scripts without headless. I have install the latest PyTorch nightly build. Wondering if anyone have the same issue and know how to solve it, thanks!

RuntimeError: nvrtc: error: invalid value for --gpu-architecture (-arch)

System info

* Isaac Sim Version: 4.5.0

* OS: Ubuntu 22.04

* GPU: RTX 5090

* CUDA: 12.8

* GPU Driver: 570.153.02

Hi, this is because cuda does not match the GPU. I used the following command to solve it. I hope it can help you!
$ pip list | grep -i cuda
nvidia-cuda-cupti-cu11 11.8.87
nvidia-cuda-cupti-cu12 12.8.90
nvidia-cuda-nvrtc-cu11 11.8.89
nvidia-cuda-nvrtc-cu12 12.8.93
nvidia-cuda-runtime-cu11 11.8.89
nvidia-cuda-runtime-cu12 12.8.90
nvidia-lula-no-cuda 0.10.1
$ pip uninstall -y nvidia-cuda-cupti-cu11 nvidia-cuda-nvrtc-cu11 nvidia-cuda-runtime-cu11

[Mming02]

Thank you very much！！！ This problem has bothered me for a long time .You make a great help to me!

[xuxuezhou]

Hi, this is because cuda does not match the GPU. I used the following command to solve it. I hope it can help you! $ pip list | grep -i cuda nvidia-cuda-cupti-cu11 11.8.87 nvidia-cuda-cupti-cu12 12.8.90 nvidia-cuda-nvrtc-cu11 11.8.89 nvidia-cuda-nvrtc-cu12 12.8.93 nvidia-cuda-runtime-cu11 11.8.89 nvidia-cuda-runtime-cu12 12.8.90 nvidia-lula-no-cuda 0.10.1 $ pip uninstall -y nvidia-cuda-cupti-cu11 nvidia-cuda-nvrtc-cu11 nvidia-cuda-runtime-cu11

Thanks for your reply! However, there’s no CUDA 11 installed on my system.

$ pip list | grep -i cuda nvidia-cuda-cupti-cu12 12.6.80 nvidia-cuda-nvrtc-cu12 12.6.77 nvidia-cuda-runtime-cu12 12.6.77

[zzzltmy]

I am facing the following error when I run scripts without headless. I have install the latest PyTorch nightly build. Wondering if anyone have the same issue and know how to solve it, thanks!
RuntimeError: nvrtc: error: invalid value for --gpu-architecture (-arch)
System info

* Isaac Sim Version: 4.5.0

* OS: Ubuntu 22.04

* GPU: RTX 5090

* CUDA: 12.8

* GPU Driver: 570.153.02

Hi, this is because cuda does not match the GPU. I used the following command to solve it. I hope it can help you! $ pip list | grep -i cuda nvidia-cuda-cupti-cu11 11.8.87 nvidia-cuda-cupti-cu12 12.8.90 nvidia-cuda-nvrtc-cu11 11.8.89 nvidia-cuda-nvrtc-cu12 12.8.93 nvidia-cuda-runtime-cu11 11.8.89 nvidia-cuda-runtime-cu12 12.8.90 nvidia-lula-no-cuda 0.10.1 $ pip uninstall -y nvidia-cuda-cupti-cu11 nvidia-cuda-nvrtc-cu11 nvidia-cuda-runtime-cu11

Thank you very much! Your method solved this problem for me. I really appreciate it, and wish you every success in your research!

[XinyuKhan]

Same here

[kellyguo11]

The blurry rendering issue should now be addressed in Isaac Sim 5.0. You can access the open source repo at https://github.com/isaac-sim/IsaacSim

[ologandavid]

Has official IsaacLab support for IssacSim 5.0 been released?

[XinyuKhan]

Has official IsaacLab support for IssacSim 5.0 been released?

5.0 can be built from source. I have tried this on my 50 series GPU.

[mi-robotics]

I am unable to get any rendering working.

following the isaacsim 5.0.0 pip install
https://isaac-sim.github.io/IsaacLab/main/source/setup/installation/pip_installation.html#installing-isaac-lab

i can launch isaac sim, create the empty scene in isaaclab and train ant, but i cannot visualize ant:

I have tried downgrading to isaacsim 4.5.0 and isaac lab 2.1.0 following these steps but i cannot launch isaacsim:

conda create -n env_isaaclab python=3.10
conda activate env_isaaclab
pip install --upgrade pip
pip install torch==2.5.1 torchvision --index-url https://download.pytorch.org/whl/cu128 ( or pip install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cu128)
pip install "isaacsim[all,extscache]==4.5.0" --extra-index-url https://pypi.nvidia.com
isaacsim 

system :

• Isaac Sim Version: 4.5.0

• OS: Ubuntu 22.04

• GPU: RTX 5090

• CUDA: 12.8

• GPU Driver: 570.169

please help, detailing your steps to get visualization it is much appreciated :)