Here is another test I am doing for a video using the archviz sample scene in UE5.0.3:
No Nanite 11.39ms
All static Meshes with Nanite Enabled via content folder
Better performance with Nanite but notice the GPU is using more watts and has a higher clock MHz? But take a look at geometry density with and without Nanite:
Without Nanite
With Nanite
It’s clear that Nanite cut down the geometric detail, so to a certain extent we are getting basic subpixel(visual) optimization like LODs but nothing revolutionary other than the fact that I can’t make LODs as fast as Nanite can(as it actually making levels of detail not performance). After trying to see if I could achieve the same(probably even better) performance without Nanite by editing the polycount of the objects manually…UE crashed.
If I’m rendering lower geometric detail with Nanite, my computing should go up but it doesn’t. It’s just using more of my GPU in comparison to regular pass. If I was rendering the same geometric detail via optimizing the meshes, I would either have higher fps or lower GPU usage.
This is a workflow issue, not a “LODs are bad” issue.
To Epic: Stop trying to force something that has worse performance as a bandaid to your workflow issues and deficiencies towards LODs.