isaac
isaac copied to clipboard
ComputationalRadiationPhysics
Lighting/AO adaptations to prevent visual artefacts
As can be seen in the above picture (indicated by the arrows) at the beginning of the simulation the structure of the GPU grid appears to be visible. This happens for particles as well as densities. As the simulation progresses the effects fade, but this takes some time and the structure suggests that there is something happening at the boundaries of the sub-domains, which does not correspond to reality.
I think if you zoom in the spheres' surfaces are clipped at the super-cell boundaries. (see noisy structure, up arrow) Additionally AO has known GPU/MPI boundary issues (left arrow). The first should be fixed, the latter only occurs with AO:
@Anton-Le Could you also provide a zoomed in picture (from your last simulation).
Here's a version later on in the simulation. It may be a tad difficult to see in the static picture, but in the animation it is apparent that there are dim joints at the indicated locations.
Particle size set to 0.3
@Anton-Le Thanks do you have a extremely zoomed in image where the clipping of the sphere is visible?
In more detail, the lines are next to clipped particles and most likely align with super cells borders (not GPU boarders).
Here's a brightened version with reduced contrast from the beginning of the simulation.
Visualized is the current density. Note the distinct lines in the domain which appear to correspond to the boundaries between GPUs
