KhronosGroup
Performance on high draw calls
The attached GLB has ~3000 draw calls, and performance on it is poor (~20FPS on a relatively powerful desktop). Three.js based viewers and babylon.js have much better performance (locked to 60FPS on the same machine). I understand that the sample viewer is meant to be a reference viewer and not compete with production level viewers in performance. My question is whether any improvement on this is planned?
I assume, that they batch a mesh (primitive and the same material) into one draw call and use internally instanced rendering. For this asset, of course, this does provide a big boost. The glTF Sample Viewer - for any asset - currently iterates the node hierachy and - if required for transparency - sorts them. Technically correct, but regarding performance of course not optimal. At this point of time, regarding performance, nothing is planned.
Understood, thank you. Regarding your assumption - it's correct for babylon.js but the three.js based viewers actually preform all the draw calls. They only use instancin if EXT_mesh_gpu_instancing is enabled for the asset.
Okay, I'll keep the ticket open. Maybe somewhere - a low hanging fruit - can improve the performance.
@UX3D-becher @UX3D-eckerlein Maybe someone can have a look, if e.g. somewhere some tasks can be avoided?