Is it worth tripling the memory requirements to hold all that vertex weight data in memory?

It depends on how well it works. I’m not sure that the memory requirements would be any greater than the triple vertex data already defined by falloff spheres.

If a joint has 3 maps, then it certainly will triple the data, versus a joint that has only one map. That's a certainty.

In your response, though you compared triple vertex data with falloff spheres, which is not the point I was asking about. I was comparing weight mapping, one per joint, versus three per joint.

But now that you mention it, I hope you do realize that the amount of data for a falloff sphere is a constant, and has nothing to do with how many vertices are affected, right? A falloff sphere is a huge data reduction in comparison to a weight map, even if the map only has a few hundred vertices in it. Which means that weight maps, even those defined per-polygon-group, are a huge memory chore compared to falloff zones.

Even capsule zones are only 40 bytes of data per zone. Whereas a 3-axis set of weight maps for a single group of vertices is 36 bytes per vertex, assuming only one joint affects the vertex. If 20 joints affect the vertex then that would be 60 maps, not 3 maps.

---

Renderosity forum reply notifications are wonky. If I read a follow-up in a thread, but I don't myself reply, then notifications no longer happen AT ALL on that thread. So if I seem to be ignoring a question, that's why. (Updated September 23, 2019)