MultiMesh without transform buffer

[dnsiltei]

I was wondering if there could be some vram bandwidth or footprint benefit in having something like a MultiMeshBufferless resource, where you would set up only a mesh, instance count and then do the transformations/color in the vert/frag shader based on the MultiMeshInstanceXDs transform or maybe even global, using VERTEX and INSTANCE_ID, with a heightmap texture, or a noise texture(textures that could be used for other things other than being only a buffer for a single MultiMeshInstance), a hash function etc...