So I am doing some arch viz, I want to optimize scene, I am doing that by reducing polygons in 3ds max from high poly to low poly model, and trying to have a good looking model. Now I have one sculpture it have 19k polygons, when It import in UE it will probably be 38k tris. Anyway is it better to leave it like mesh, or this flat piece of model convert to some displacement? I have feeling that displacement is good when you hold number of tris low but when you try to look nice it bring too much polygons/tris.
Second thing about optimization is LOD. I used to do it on UE4 and it was pain. I can still do but my friend told me that UE is doing that automatically? Also in UE4 there is distance to adjust where to switch from one LOD to another, it is visible all time, friend told me they made blend in UE5 and its not visible? I am little confused new UE5 can hold really big polygon count, but I still believe lower lever is better for FPS and working Real Time in general. Can somebody direct me and tell me what is best approach for optimization. Displacement vs polycount etc. I am looking for realistic RT scenes so quality is must. Also i didn’t put this to arch because it is low flow of people and question is related to games also not just arch viz.
Thanks 
38k tris is not too much for archviz. 380k is not so good. 3800k is overkill. So, forget about this “problem”. Displacement could even harm performance.
Automatic LODs creation in UE5 is effortless and very efficient. Then you can adjust it – also very easy.
My whole apartment have less than 3 M polygons (this that I am working on), witch is good I think. I replaced some high poly models with lower ones. Generally my apartments have 5-10 M polygons. I know Nanite can handle billions of polygons, but what about smartphones, tablets, older GPU-s like 1080 or 970 GTX cards that doesn’t have RTX etc (this is for real time walkthrough i will pack as .exe and send to client). Is there cheat sheet for this or something so we can align with. I was thinking this model to use on PCVR but now that is under 3M polygons I think it can go on standalone VR (this is for my presentation)? I need some orient on this so I know where limits are or it all goes on try and error learning curve? So conclusion Is I know Nanite can handle big amount of polygons but what about things that doesn’t support it, displacement is also not an option, looks like your only choice is to put down polygons by big time and find balance so models look decent?
There is no, 'coz polygon count is not everything that affects performance. That’s what I’d do:
- Add auto-LODs. Adjust LODs settings
- Add some lights (new megalights give +5-10% fps), adjust lights quality
- Use Lumen GI without Nanites, adjust settings in PostProcess volume. Use baked lighting if there is no other way (but this way you can have +100-300% fps).
- Add quality switching option (Low, Medium, Epic, etc.)
- Package project
- Test project on random hardware
- If fps are not good, repeat from 1
with megalights it went from 35 to 110 fps 
I want quality I will bake it on the end