[question] Multi-GPU setup fails on DepthMap Estimation

Describe the problem

Tried on 4 different setups running Linux. Single-GPU setups all worked, Multi-GPU setups all failed in trying to estimate the Depthmaps in a standard photogrammetry pipeline... added also an issue to the alicevision repo...

Dataset Any set of images...

Log

https://github.com/alicevision/AliceVision/issues/1514

Desktop (please complete the following and other pertinent information):

• OS: Linux Ubuntu 22.04.3, Cuda 12.2, latest 535 nvidia drivers
• Python version 3.10.12
• Qt/PySide version 5.15.3+dfsg-2ubuntu0.2
• Meshroom version:
  • Binary version 2023.2.0

Additional context I believe it is the alicevision pipeline that has some trouble, but I am not sure and did not receive any feedback there...