Right, vertex works but depending on the mesh it can be several times as memory heavy than point for the same result as there is a vertex for each primitive a point is connected to.
As a test, make a 1000x1000 grid and give it a color. If you store it on points and export it as a FBX it’s 73 mb. If you store it on verts, it’s 134 mb. Unnecessary waste.
Vertexdata in houdini can be looked at as smoothing groups if you come from another package and should really only be used to define hard and soft normals (at least when we are talking about things that eventually will become FBX). For other data, points is the better option.
As for the grid part, I’m not against having that topology. It’s the creation method I’m questioning. If you add noise to the mesh in the picture, it’ll break early. Small amounts are ok but you’d need more topology for any significant distortion. That’s where the problem comes. You can’t change the topology in this setup as you have based the ramps on the pointnumber distribution. A more stable way would be to sweep a 2x2 grid along a resampled curve. It would 1, give you the ability change the topology and density. 2. You have more control over a curve than a twist deformer and 3. It automatically calculates curveU based on the curve length so you don’t need to do that manually.
You could also base the curves on the UV coordinates so you can have a ramp (or three…) going across as well, allowing you to fade the hard edges out in case you ever want to use a tiling texture.