InstancedMesh - does it require a specific feature that not all GPUs have?

Hi all,
On one computer InstancedMesh improves performance, on another it actually drags performance down.
I’m trying to figure out if this is because I draw more faces with InstancedMesh, because the bounding sphere is larger (to include more meshes) so culling is less accurate / I draw some stuff outside screen, or is it because on the weaker computer I test on, which is a laptop, instancing is not properly supported and I suffer the consequences of a slower fallback being used internally?

How do I validate / check this?

1 Like

There is a thread here but i can’t find it now that catches this on some browsers on some devices. It really shouldn’t be dragging the performance down, but you might be running into this specific issue.

ANGLE_instanced_arrays is the required WebGL extension, but I think it’s pretty universally supported at this point. I would expect a hard failure if it’s not supported, rather than just degraded performance.

I don’t think InstancedMesh supports front-to-back sorting, like separate objects would have? Without that, it would be pretty easy to stumble into situations with bad overdraw using instancing, so be careful of that.

Also see: Merge or Use instance and Native render make no difference with Chrome plus Integrated Graphics Card


Ahh good point on the overdraw, you could hit some pretty bad cases if you start filling a lot of pixels back to front.

Interesting, thanks!