Replying to @snowboyken
You are going to have a hard time processing tiles on a 3060TI.
Just make triple sure to close the minimap and/or just manually import one tile at a time.
I would also suggest just staying away from the engine as much as possible for your scope.
Use Houdini to build things - then eventually import into unreal.
Replying to anon79855654
Almost all of them - actually all the ones I know of except two - are a custom take-down that has been specifically gutted from source.
The one that’s not barely uses landscapes at all. The other is mentioned below.
Not a single AAA title I have played uses the actual landscape system in release.
ARK did. Look at what a pile of hot steaming s*it Wildcard made of that! (Not to mention that nearly 6 years down the line they have the same exact game crippling bugs which are wholly due to the engine itself, inadequacy of developers, and lack of leadership/testing). [fun fact on this since i’m re-posting nearly a year later; This month (October 2023) a new version of Ark was released useing Unreal 5. It’s the exact horrible non-working 2 frame an hour mess on a 4090 you’d expect out of Wildcard! Quite amazing. Years pass, ZERO learning from the devs. you joke, but it takes some SERIOUS skill. Or at least “consistency” just in a very, very WRONG way :P]
There really aren’t, unfortunately.
There’s teams who decided to put time into gutting the source to get something going OK-ish.
And there’s smarter people who avoid the engine like the plague it is.
That’s about it.
But obviously if you think what you wrote (and based on how you wrote it. Really, what do you do with all the time you save writing U instead of you? Think up BS for the next post?), you haven’t played any unreal made game at 4k, so your opinion is about as useful to the performance conversation as that of the idiot who owns facebook was to congress…
Original posts have been removed from this and the whole topic which was already a mess full of disinfomration, hearsay, an BS, tough more legible is now missing some rather relevant parts. Since the offending bs from at least a banned user and another one are gone, I’ll re-word the original messages to re-provide the answers that were removed.
Ideal tile Size:
Anything Epic or anyone would say on the matter is purely Anecdotal. 100% situational. And wholly unneeded.
As such, it should never be included in any documentation, or matter one bit.
The only way to find what tile size works for X project, is to actually Test X project and bench the baseline worse possible situation.
In other words, Do the work.
For world partion (the older system), All it does is it loads levels (via streaming).
You can set it to load a maximummof 4 levels simply because your player can stand at an intersection of 4 maps.
Your worse possible performance will always include 4 levels.*
That’s not infinite.
That’s not light on performance because each landscape tile comes with x*x components worth of draw-calls.
And in 20223 that definitely isn’t the proper way to load things in (it wasn’t in 2000 so why would it be now?).
Curiously, it’s now nearly a year and a half since those posts I think.
If you look at the direction that landscape has taken in relation to my post on August 22 (UE5 - World Partition - Massive Map Sizes? - #40 by MostHost_LA)
-
It’s still the same exact mess, but queue in Nanite, which does exactly this.
-
Nanite does this - badly still - but a lot more efficiently than the landscape itself.
-
Still non-existent with nanite, but if you replace all of your landscapes out to heightfield meshes you can actually achieve something very similar without needing Voxels.
(costs too much performance for 4k rendering though. so usual Bench and Give/Take).
-
Non-existent - ignored.
BUT - since nanite is a single mesh running a single material, this is at least in theory Possible now.
the problem there is that you can do sphere approximations, not ESRI derived values
-
Actually I have no idea what Nanite is doing in terms of converting the 1 point per meter fatally flawed landscape system under the hood into it’s final mesh representation - BUT - I know it’s using analytical generation, and it’s therefore bound to provide a rather drastic overall tris reduction.
It’s interesting that they expressly chose to build on top of the dead elephant.
But it’s also nice that they are at least attempting to address the so called (dead) elephant in the room…
As of this writing, Nanite for landscape is no-where near production ready: Keep making your own Mesh Only landscapes instead…
The idea though, is that Eventually at least, it won’t matter if you utilize meshes, or if you stuck it out for 2 years with a landscape and 4 FPS from it. Both Mesh and Landscape alike should be able to achieve the exact same performance overall by switching Nanite on.
Let’s also keep in mind that these considerations are for RENDERING only.
Ram and CPU are still very much a thing that you’ll need to profile for even with Nanite taking over/the load off/ the GPU when it comes to the landscape.
So, same answer as before; Do The Work…
