GPU memory allocation in Vulkan seems to be a huge headache.
That seems to be the underlying cause of this ray tracing problem. It's exposed to the higher levels, but in strange ways.
I use the Rend3->wgpu->Vulkan chain, and can't get any info about how much GPU memory is available or left. If I load too many textures into an NVidia 3070, they spill into main memory and rendering continues, slowly. If I load too many meshes, I get a fatal error from the Rend3->WGPU layer, because the allocation that failed happens long after I made the request to load a mesh and it's too late to just return an error for the request.
Can anyone explain why it's so hard to find out the memory allocation situation in Vulkan?