Setup: Windows 11 + WSL2, RTX 4060 Laptop GPU, driver 581.32/CUDA 13.0. Docker (native in WSL2, not Desktop) + NVIDIA Container Toolkit 1.19.1, --gpus all. ParaView 6.0.0 official Linux binary, confirmed via pvpython --version. EGL vendor ICD already set up (/usr/share/glvnd/egl_vendor.d/10_nvidia.json → libEGL_nvidia.so.0).
What works: nvidia-smi inside the container, nvidia/cuda test containers, and VisRTX — which correctly logs VisRTX 0.1.6, using devices: 0: NVIDIA GeForce RTX 4060 Laptop GPU at startup, in the same process.
What doesn’t: A plain pv_simple.CreateView("RenderView") + SetOffScreenRendering(1) (library import, not pvpython CLI) defaults to attempting GLX (vtkXOpenGLRenderWindow ... bad X server connection). Forcing VTK_DEFAULT_OPENGL_WINDOW=vtkEGLRenderWindow removes that warning, but nvidia-smi stays flat at 0% GPU-util/0MiB/no processes throughout, even while actively rotating a loaded mesh — tested across several env-var combos with the same result.
Already checked: This looked like the bug fixed in this thread (VTK_USE_X=OFF letting OSMesa clobber EGL in the CMake factory override), but I diffed Rendering/OpenGL2/CMakeLists.txt at the exact VTK commits pinned by ParaView 6.0.0 and 6.0.1 — the fix is already present in both, so it’s not a recurrence of that one. Also confirmed no d3d12_dri.so in the image, ruling out the WSLg/Mesa-D3D12 path as an alternative.
Question: Given the CMake-level fix is already in, is there a known reason vtkEGLRenderWindow init would still silently fail specifically under WSL2 GPU virtualization vs. native Linux? Is there a way to get non-silent diagnostics from the EGL init path? Happy to run whatever’s useful.