odf posted at 7:21 PM Thu, 9 June 2022 - #4439761

Out of interest, do you happen to have a good source that explains why? I have a distinct memory of having watched a whole video course about light transport at some point, but apparently my scatterbrain has not held on to much more of it than "we can go a lot faster than Monte Carlo by picking the rays we trace back smartly." :-)

Some of the best sources on it are from Epic Games, Intel, and Nvidia. I don't have specific links. Tessellation is very similar in how it works, and why it takes so long in a path tracer.

The new Unreal 5 engine is a prime example of it of why micro displacement is a really bad and outdated idea. Unreal 5 can handle millions of triangles in view at the same time, eliminating the need for micro poly anything. All while rendering far better and far faster than most engines. Unreal 5 is a path tracer that will probably never get support micro displacement. There is no need for it, and there is a good reason why. If you do need to use it (obviously some Poser content required it), Poser still has Firefly, Studio still has 3Delight, etc.

Micro polygon displacement can be controlled a few different ways. One way is how it is done in Blender with an adaptive node that modifies the geometry before or as it is sent to the engine, each time, more on this later... Another is by assigning a micro polygon displacement rate and ratios, and doing it during the initial render setup like FireFly does.

As example, Luxology uses rate and ratio to stop you from killing the render times. The settings were exposed in Luxology, and you could easily kill render times, run out of memory, or just crash the system. Ludology is also a Monte Carlo based engine that uses irradiance caching very similar to Firelfy, both are hybrid engines. Basically the settings are a cap to how far it can go before it causes problems, even in a Monte Carlo engine.

Both ways have limits to what they can do, and how much detail they can show before memory issues spiral out of control. 

All of this doesn't even get into setting the scene up multiple times in a path tracer, because where it looks from changes every bounce, on every path. That is where it kills a path tracer, changing views on every bounce, for millions or billions of paths. Monte Carlo paths are basically limited to the number of pixels, times the number of bounces. And it only has to set the scene up based on the camera view, and not every bounce.

---

Some things are easy to explain, other things are not........ <- Store ->   <-Freebies->