Unreal Engine Geometry Scripting Procedural Terrain Test
Flythrough of a very basic but very large procedural desert using the experimental Geometry Scripting plugin in UE5. Some notes:
-Total terrain size is 2500 square km.
-Terrain is rendered with a 5km view distance.
-Terrain is broken up into 10K chunks at 2.5km square apiece, building 8 chunks per frame.
-Terrain is built using Fast Noise Generator, sampling two noise layers, one perlin & one distance cellular.
-Total time to generate is around 1 minute. (As of this upload, DateTime comparisons in BP are broken and so I can't get exact timing metrics on my machine.)
-Each 2.5km square chunk tile is tessellated at a factor of 20, or 256 tris.
-Collisions are not being generated, this greatly increases per-chunk generation cost & slows down the generation process.
-Normals are generated per-chunk, meaning there are seams. I need to look into manually building normals per-tri using the noisefield data.
Overall thoughts: Geometry scripting is stupidly powerful for a blueprint workflow, with the proper precautions. I'm having to manually step through my chunk array, deferring portions of the work across frames to keep things from locking up/overloading, and I need to see how performance goes with manual normal calculations being added, as well as generating collisions only on chunks relative to gameplay, but these things are very doable with more time.
