I’m testing the SNES solver for 3D cases, and I noticed a significant slowdown even when using only a few points in the toroidal direction. The 2D cases work really well, so I am trying to understand what changes when moving to a minimal 3D setup.
For example, I used:
MZ = 5
zperiod = 1000000 # I tried zperiod = 1 and it seems to work somewhat better.
[mesh:ddz]
first = C2
second = C2
upwind = W3
The case is still much slower than expected compared with the corresponding 2D setup. Does anyone have an idea what might be causing this? Could this be related to how the Jacobian/coloring is constructed in 3D? Or perhaps with the Z periodic BC? where is the BC in z applied?
I’m testing the SNES solver for 3D cases, and I noticed a significant slowdown even when using only a few points in the toroidal direction. The 2D cases work really well, so I am trying to understand what changes when moving to a minimal 3D setup.
For example, I used:
The case is still much slower than expected compared with the corresponding 2D setup. Does anyone have an idea what might be causing this? Could this be related to how the Jacobian/coloring is constructed in 3D? Or perhaps with the Z periodic BC? where is the BC in z applied?