Nanite Performance is Not Better than Overdraw Focused LODs [TEST RESULTS]. Epic's Documentation is Dangering Optimization.

The reason I did the test was to disprove theories and official lessons by Epic about poly density helping performance where culling isn’t happening all the time. And to prove the 5th iteration of Nanite still doesn’t include substantial performance gains like everyone claimed would happen a couple of years ago.

(Nanite needs a substantial change in how it works to improve performance. We’re talking such massive changes to the point where calling it “Nanite” over “Nanite 2.0 etc” would be a bit of a cheat)

@Lucian The reason why your not seeing that much of a performance difference is because you started off with much more nanite overdraw. A scenario where you would be told to follow “protocol” and try and get more clusters.

You have to do the same thing when you compare quad overdraw with Nanite. You have to analyze the surface area of heat in your frame.

I’ve already ready proven that:

More quad overdraw surface area in frame=more gain with Nanite or x2 the Nanite gain with proper LODs and batching.

So with your test it’s:

More Nanite overdraw=The less gains micro variance displacement/clusters will help performance.
(since flat subdivision to hack in clusters gets compressed to zero clusters with Nanite)

You’re getting the same amount of overdraw in different forms. In your “low poly”, Nanite’s pixel overdraw is spread across the screen. With displacement/clusters, overdraw in getting packed in the same pixels over & over which doesn’t allow you to measure the situation correctly unless you know how this stuff works. Again, this is why the view mode is so ineffective.

(Again, your welcome Epic, invest in fixing this)

So lets measurement performance scenarios from best to worse for anyone wondering:
(I’ll remind everyone here that non-nanite meshes also have to watch pixel overdraw. It just doesn’t have a viewmode in unreal like Nanite does)

Each dropping significantly in performance potential.

  1. Non-Nanite, Quad Ovedraw and Pixel Overdraw contained scenes
  2. Nanite scenes with low Pixel Overdraw
  3. Nanite scenes with high Pixel Overdraw
  4. Non-Nanite, Quad Ovedraw and Pixel Overdraw overwhelmed scenes

So basically you have to neglect optimization (again) to gain with Epic’s suggestions.

I will link your test and my reply of context to the main mega thread.
And for anyone complaining about using “5.5 preview”, it’s the same in case with UE5.4 and all the other version since barely anything was done to improve/modify Nanite other than making other things compatible with it. I assumed this was the case and went back to prove myself right (again).