[Sionna/Dr.Jit] OPTIX_ERROR_INTERNAL_COMPILER_ERROR (7299) during load_scene

[shamsul-haq-shuvo]

I am encountering a crash when trying to run Sionna RT using the load_scene function. The error appears to be an internal OptiX compiler error coming from Dr.Jit/Mitsuba.
While the Mitsuba variant cuda_ad_rgb sets correctly, executing load_scene("first.xml") immediately terminates the kernel with an OPTIX_ERROR_INTERNAL_COMPILER_ERROR (Code 7299). The error log points to a failure in optix_core.cpp within the Dr.Jit backend, despite the environment seemingly detecting the GPU correctly. I am running Python 3.10 and would appreciate any insights on whether this points to a specific NVIDIA driver incompatibility or a corrupted Dr.Jit installation.

OS: Linux / WSL2

Python Version: 3.10

Environment Manager: Conda (sionna_fresh)

GPU: RTX 5060

NVIDIA Driver Version: 581.80 (Tried with latest version to. Same problem)

[merlinND]

Hello @shamsul-haq-shuvo,

Could you please try without setting the Mitsuba variant?
Sionna sets the appropriate Mitsuba variant automatically upon import (cuda_ad_mono_polarized, not cuda_ad_rgb).

[shamsul-haq-shuvo]

Tried. Nothing works. I have been trying everyway possible for the last 4 days. Couldn't find a solution.

[merlinND]

The internal compiler error comes from OptiX, however you mentioned already trying the latest driver so I am now sure how it can be fixed.
The issue is that WSL2 is not officially supported, so it's not even really supposed to work. I will link previous discussions about Sionna / Mitsuba on WSL in case it helps:

• GPU based ray tracing not supported on Native Windows and WSL #593
• https://mitsuba.readthedocs.io/en/latest/src/optix_setup.html

I don't know whether using Conda interferes with anything, might be worth trying a plain Python / Pip setup.

[shamsul-haq-shuvo]

I tried the previous discussion way yesterday. Didn't work.
If I dualboot my pc with linux will it work?
I didn't find any other workaround.

[merlinND]

Yes, Linux should definitely work!

[LucasVsu]

hola, ese problema me paso igual en wsl2 y docker. Linux nativo (ubuntu ltsc) y Google Colab, sin problemas!