Face Centered Cube Lattice

[Meerkov]

I'm wondering if you've read https://www.researchgate.net/publication/26659194_Efficient_LBM_Visual_Simulation_on_Face-Centered_Cubic_Lattices

This 2009 paper shows that D2Q7 hex converges to the same result as D2Q9, with seemingly a 50% speedup. It seems to me that this speedup could be an illusion due to perhaps unoptimized code, however it does seem plausible that using only 78% of the memory should result in a speedup.

They similarly test the FCC D3Q13 lattice against D3Q19. The FCC lattice seems to need MRT, but otherwise behaves quite well and would only need 68% of the memory. Considering FluidX3D is memory bound not calculation bound, do you think that FCC lattice would offer an improvement?