Since the latest patch we were told memory would be more accurate.
It’s been the worst it’s even been, we did more than 10 calculations today.
1st one was 107,000
For the second one we optimized a bunch of assets, reducing the total vertex count of the map by more than 2.5 million.
2nd one was 108,000
We did this again and optimized more assets and the result for the third one was 110,000
We continued deleting assets and the last calculation gave us a result of 118,000.
Edit:
We just did another one WITHOUT CHANGING ANYTHING, and it went up to 119,500.
We have multiple questions:
1.- HOW is memory calculated, what’s expensive? Is it vertex count? Texture size? There’s literally 0 information on this, making it impossible to know what’s happening.
Do things in the Content Drawer contribute to memory?
2.- WHY is there still a memory limit if each UEFN island is on its own server? My understanding of 1.0 was that since more than one island had to be open at the same time in a hub, there had to be a limit, why now?
3.- WHAT are we supposed to do with memory varying so much without any changes?
I am having the same issue. I have been using the message log to check the usage of my custom assets but even after fully deleting the assets the memory will increase.
On top of this, the memory overlap is extremely bad between grids and it doesn’t seem like there is a way to override the distance an object is rendered.
It would be extremely useful to see the effects an object has on the surrounding area, not sure why this was removed.
Aside from these obvious issues there is definitely an inconsistency between memory calculations. I have been running calculations today when submitting my map and my memory has once again randomly increased by 20k despite no changes being made between submissions and every calculation prior showing a much lower and more consistent amount.
I had similar issues while doing some testing of World Partition specifically (which I think is causing inconsistencies); placing large groups of meshes far apart from one another without any landscape. It became evident that either world partition distance settings aren’t being respected, the documentation for ‘These things are always loaded, these things load based on your location’ is wrong, or the entire measurement system is just outright busted.
You can reproduce pretty easily by just placing some prefabs a long distance apart - the memory barely budges as you move around, and will skew 5-10% with each run.
This is my theory also. I am working on a map that heavily relies on world partition and I could be wrong but I believe the project Rynex is referring to does also. I have noticed that smaller scale maps that do not use world partition made in UEFN have not been affected as bad by these changes.
Inconsistencies between memory calculation are being investigated and some fixes are already in the pipe. Incoming releases should adress some of them.
Vertex count, animations, audio…etc. are all contributing, but often the biggest cost is textures.
e.g. using 8k textures when only sub mip-map are really used → the all texture is still in memory.
Note that the memory reported (bar at the top when in edit mode in FN) is different in islands built with UEFN vs islands build with Creative. This is because the way objects are streamed in the scene is a bit different.
Note also that the memory calculated and reported in the bar only relates to runtime memory consumption.
i.e. Assets that are in the content browser but never used in the world shouldn’t affect the memory consumption.
After doing multiple tests, I realized the Loading Range value in the World Partition Streaming panel is not taken into account. Only Cell Size. Like @skyzyn mentioned in this post, and @defcon117 did too in another post.
Is this something that’s being investigated?
it seems it’s fixed at 25600 units which is the default it comes set at.
It doesn’t matter if you set it to 1000 or 500,000, it will always load/unload at 25600 units distance.
The same map had a weight of 102,000 and by deleting things has increased. The truth is that I can not continue if I do not understand how to comply with the size operation.
