WORLD COMPOSITION Retains Ram Load Even After Tiles Have Been Unloaded. Can I get some help?

Hello, hello.

We’ve discovered a pretty debilitating issue with World Composition when running some tests for our project. We added 400 tiles (1km x 1km per tile) into the engine only to find that if all tiles are loaded up together, they use roughly 5+ gigs worth of memory (Although this number isn’t stationary. The longer the engine runs the higher the memory usage climbs).

This is of course pretty standard. The problem arises when we attempt to unload the tiles. Meaning if we hide the remaining 399 tiles and only leave one 1x1 tile to remain on the screen, the memory usage remains the same (5+ gigs of RAM usage). Is this normal? Is this the way World Composition is supposed to work or are we missing something when it comes to optimizing the usage of tiles and World Composition. Any help would be appreciated.

We’re pretty desperate in looking for a solution to this issue. We’ll probably be bumping this once a week in order to see if there’s anyone out there with the experience we need to help us figure this out. Thank you. Will add pictures of this issue along with the RAM usage numbers shortly.

Are you doing this RAM usage test in the editor or in a cooked final build?

No. I’m going to try building the project and seeing how it works.

No, I don’t think thats how it is supposed to work, I’m almost positive.

Assuming I’m right, though I"ve not been able to use it yet, given I have no qualifying terrain ( working on it now), the memory used would likely be , whatever size the terrain is, + objects + textures , lights you name it, that you have on those segments, in addition to however many ‘segments’ are loaded , surrounding the player, ie: paged terrain , which is what this does.

Are you manually doing this as opposed to letting the paged terrain class take care of it , if so why, just wondering.

It sounds like a memory leak to me but then I’ve not used it enough to say, sorry.

I suspect it is, or you’ve done something maually wrong.

The engine is supposed to take care of this automagically.

The idea of paging terrain is to make HUGE landscapes and manage memory/cpu usage by unload/loading, as you know, terrain segments.

It should not increase over time ( unless some zones are MUCH larger/have more content), but someone more knowledgable than me shall have to weigh in. You should see some consistency, say with zones that have large objects , vs ones that are mainly just plains of grass, etc. Thats the idea anyway.

Good luck with it.
nl

That’s the problem. There is no performance gain when far tiles (or all tiles) are unloaded.

The longer the engine runs the higher the memory usage climbs < Sure sounds like a memory leak as I noted.

The problem arises when we attempt to unload the tiles < here he didn’t say outright, if its just ‘5+’ or also ‘continues to build as it goes’.

Not having used it on such a large terrain I have no idea, but I can’t imagine its escaped every dev’s eye after so long, but maybe few have huge terrains, dunno.

I imagine the kite demo was huge enough to have seen some form of issue, and indeed Ive seen people say it takes AGES to even load the data, so that tells me, many more should be commenting about out of control memory spikes, etc. We aren’t seeing that, faik.

Hope you resolve it, its going to take someone far more knowledgeable who has used WC , or OP finding a possible mistake of use on his end.

Cheers

That’s why I asked if he’d tested with a cooked build. Caching and not releasing resources may be a behavior of the Editor, not the Engine.

This is as far as I know, true. Work for hours in editor, raise up your ram usage to 8.000, open a completely empty world. You will see ram usage stays same until you restart editor.

On the other hand this behavior does not happen on a packaged game.

If a thread discussed this I either missed it or not much has been said about this ?

Is it only happening on 4.11.previewX ?

thx