Good day everyone!

A few weeks ago I already tried to ask after this issue, but there were no answers unfortunately.

My question is:
- In Solaris, rendering with Karma, why dicing based adaptive subdivision increases vRAM / system RAM usage till rendering the final scene VS other render engines, where this increase in RAM usage is not exists?

As an example:
Blender Cycles, adaptive subdive, dicing: 1 - used RAM: 8Gig
VS
Houdini Karma (mesh threated as subdiv surface) dicing 1 - uses nearly 30 Gig ram (if dicing disabled / turned down to zero / no displacement is used, mesh is not threated as subdiv mesh - the RAM usage is still higher vs other engines, but it is close to them - I also have to add: in this scenario Karma is the fastest, but as soon subdiv jumps in, it will be the slowest - XPU, so logically it runs out of my Vram)

Same scene, same camera distance, same lights

Could it be that I'm missing adaptive hidden setup somewhere? And Karma is subdividing the mesh as the camera would be at the closest clipping distance? Or based on UDIM res, it ignores the camera distance and playing with dicing 1 based on UDIM resolution?

Thanks in advanced if you can help!