Could you check when sensorsim image support for Blackwell ?

[chocwjc]

Is your feature request related to a problem? Please describe.

I have RTX Pro 6000. But sensorsim container can not working on it.

Describe the solution you'd like

Could you check when sensorsim image support for Blackwell ?
And where can I download the latest one?
https://forums.developer.nvidia.com/t/questions-about-nurec-container/350490
In here, "The nurec container itself is not open source—it is a pre-packaged runtime you pull from NVIDIA’s registry (NGC), including proprietary checkpoints. Its code is closed, although you can use its models and API for your own data."
However, I can not find "nurec container" on NVIDIA’s NGC.
And the "https://hub.docker.com/r/carlasimulator/nvidia-nurec-grpc" is too old.

Describe alternatives you've considered

Could you check when sensorsim image support for Blackwell ?
And
Where can I downloads the sensorsim image ?

Additional context

Can I build the sensorsim render (such as libnrend_cc_100.so) ?

[annz96]

But sensorsim container can not working on it.
Do you have any error log for check that? Thanks.

[chocwjc]

As same as #6
And on sensorsim container, there are only libnrend_cc_8X.so libnrend_cc_90.so
Maybe need libnrend_cc_100.so

[maximilianigl]

This is the container: docker.io/carlasimulator/nvidia-nurec-grpc:0.2.0
It was initially released for Carla, hence it's hosted there.
Unfortunately I can't give a timeline regarding the next release, but it's actively being worked on.

[frrrizzz]

sm_120/RTX 5090-Support for the NuRec-Container would be required. Currently, the NuRec-Container is limited to sm_90 capability.

[im-hashim]

• me too. similar problem

System Spec:
Ubuntu 22.04 LTS
RTX PRO 6000 Workstation Edition
Compute Capability: 12.0

Also attempted to run on
Ubuntu 22.04 LTS
RTX 2080 Ti, whose
Compute Capability: 7.5 (excepted some backward compatibility and attempted to run)

will be waiting for new container image.

[adityac-ymsli]

My current conifguration is:
Ubuntu 22.04 LTS
RTX 5090

I also am facing the same issue when I try to execute: uv run alpasim_wizard +deploy=local wizard.log_dir=$PWD/tutorial

Output error:
sensorsim-0-1 | RuntimeError: CUDA error: no kernel image is available for execution on the device
sensorsim-0-1 | CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.
sensorsim-0-1 | For debugging consider passing CUDA_LAUNCH_BLOCKING=1

[pranavdoma]

Hi @maximilianigl

I am facing similar issue running alpasim wizard on Blackwell RTX PRO 6000, Driver Version: 580.126.09 CUDA Version: 13.0.

CUDA error: no kernel image is available for execution on the device
tinycudann was built for lower compute capability (90)
system compute capability = 120

The crash originates in:

sensorsim-0 (nvidia-nurec-grpc container)

During:

get_available_cameras → load renderer → CUDA kernel launch.

nvidia-nurec-grpc:0.2.0 does not support your GPU architecture (sm_120), causing CUDA kernel launch failure.

Can you please let know if there could be a precompiled version for latest setup?

Thanks,
Pranav Doma.

[maximilianigl]

The sensorsim images are created by a different team, so I don't have perfect information here. But AFAIK the image should come any day now, with it currently going through some review process.

[maximilianigl]

Could you try this PR please: #63
it updtes to the newest NuRec image

[ikutasama]

Could you confirm that which NuRec image now supports Blackwell? e.g. RTX 5090

[im-hashim]

Could you confirm that which NuRec image now supports Blackwell? e.g. RTX 5090

I tested on blackwell using:
https://catalog.ngc.nvidia.com/orgs/nvidia/nre/containers/nre-ga/26.02.00

latest seems to be 26.04.00