HierarchicalInstancedStaticMesh Visibility Culling not working?

Can anyone confirm / does anyone know why culling isn’t actually working with the individual instances?

I need a custom “something” that will function like the lanscape foliage tool for instancing stuff.

Using the HirerchicalInstancedStaticMesh component cut performance cost in over half, but to get above the minimum I can’t have all 1089 instances render all the time.

I’m using the instance+meshes as a tiled landscape. If I look up in the sky why are those instances still rendered? What? -_-

Check your bounds scale, If the camera frustum is still in view of the bounds when looking to the sky the HISM will still be rendered.

I wish it were that easy, but you may be onto something.
the individual instances do not display any bound of any sort when you enable view > visibility bounds.

I think the overall bounds of the actor are being used instead of individual components…
how do I get this to work on individual instances?

I have some more information on this.
It seems that the instances are almost completely ignoring frustum culling or behaving as groups.

This one is particularly interesting as the blocks within the hole should be entirely invisible.

And here it is even more evident that individual instances do not take frustum culling into account. These blocks are all under the landscape but not occluded:

Any ideas on how to force culling on a per/instance basis welcome…

Also, the InstancedStaticMeshComponent doesn’t occlude at all:

One more screenshot. the Bounds clearly display the difference of how instances are collected between the foliage on the landscape and the HISM component…

Getting somewhere, but the results of the HISM component fustrum culling are still very random.
I wish I could ramp up the scale or settings for it just for the component…

The initial issue seems to have been with the mesh itself, however changing back to it afterwards had the HLOD system still occluding, so I’m not sure what I’m dealing with still.
Fact is, even with a larger mesh the occlusion from frustum culling still seems highly inaccurate.