CUDA and driver version

What is the recommended driver and cuda version for the latest release. I have a RTX 3070 and been using CUDA 11.6 and driver 510.54. I get the following error when gpu pipeline is enabled and have visualization issue ()

[Error] [carb.gym.plugin] Gym cuda error: invalid resource handle: …/…/…/source/plugins/carb/gym/impl/Gym/GymPhysX.cpp: 3417
[Error] [carb.gym.plugin] Gym cuda error: invalid resource handle: …/…/…/source/plugins/carb/gym/impl/Gym/GymPhysX.cpp: 5620
[Error] [carb.gym.plugin] Gym cuda error: invalid argument: …/…/…/source/plugins/carb/gym/impl/Gym/GymPhysX.cpp: 3442

Hi @robot_overlord,

That combination of GPU and driver should work fine. Are you able to run any of the examples? Are you using the conda setup recommended in the install instructions? Lastly, can you confirm that everything is working correctly with your driver otherwise? IE: working GPU acceleration on the desktop, nvidia-smi, vulkan-info, vk-cube working normally?

Take care,

Hi @gstate ,

I figured out a simple solution. Using 470.103.1 and cuda 11.4 seems to have fixed it. Not sure what was the issue earlier.


1 Like

I can confirm this bug as I encounter the same problem using CUDA 11.6 and driver 510.54. It would be great if you can provide support for the newest Cuda and driver version.


I can run the examples perfectly using a RTX 3080 and
NVIDIA-SMI 510.54 Driver Version: 510.54 CUDA Version: 11.6
I have PyTorch version 1.11.0+cu113

1 Like

Hi @robot_simulation ,

Do nvidia-smi , vulkan-info , and vk-cube all work for you?

What GPU are you using, and what error message are you getting?

Are you using the Conda environment setup that we provide, or something else?

Take care,

Hi, I don’t have vulkan-info , and vk-cube installed in my system.
nvidia-smi works fine.
GPUs are: 2 X NVIDIA GeForce RTX 3090
I’m using conda python 3.8 pytorch 1.10.0 py3.8_cuda11.3_cudnn8.2.0_0
Actually, the headless mode works, the problem I have is that I cannot use issacgym with GUI mode, there is no error message that I can check, because whenever I run an example code, the whole system will log out, and when I login again, no thread was there.

This error is because of installing Nvidia driver both using Debian package and .run file. So the system messes up with different versions. Installing the newest Nvidia driver solved the problem.

1 Like