Strange results with the newest NVidia driver (v531)

[sapo17]

System information:

OS: Windows-10
CPU: Intel64 Family 6 Model 158 Stepping 10, GenuineIntel
GPU: NVIDIA GeForce RTX 2060
Python: 3.9.13 (tags/v3.9.13:6de2ca5, May 17 2022, 16:36:42) [MSC v.1929 64 bit (AMD64)]
NVidia driver: 528.49
CUDA: 11.8.89
LLVM: 15.-1.-1

Dr.Jit: 0.4.1
Mitsuba: 3.2.1
Is custom build? False
Compiled with: MSVC 19.34.31942.0
Variants:
scalar_rgb
scalar_spectral
cuda_ad_rgb
llvm_ad_rgb

Description

Hello, so this is not really a question, but rather something that I encountered, and I thought of sharing it.

The other day I upgraded my NVidia driver to the newest version (531.31). After that, some of the material optimization test cases that I had written for my thesis failed (using the cuda variant). I will go not into detail about them, but please let me know if you need further information.

If someone else has a familiar problem, changing the NVidia version to a previous version might help. In my case, I downgraded to 528.49, and now my test cases work as previously.

[njroussel]

Hi @sapo17

What do you mean by your optimizations "failed"? Did they crash or produce wrong results?

[sapo17]

Hello,
the optimization procedure looks like this, but my test cases were different (using adifferent scenes and material parameters).
They produced wrong results. Although once in a while I experience a similar error like this #587 (but this is an independent topic).

[njroussel]

Ok, thank you.

Unfortunately, v531 hasn't been released on Linux yet - I cannot test this for myself. I will keep this issue open.