Okay 5.7 preview for Single Player with FastGeo is insane in Open World.
Infinite Quality Forest with 0 stutters . Still have no idea how to setup assemblies 
EPIC + Base Settings from 5.7 - full Nanite runtime 8km+ distance visible
I approve 5.7
Nanite production edition.
So Ladies and Gentelmen
-Open World 16x16 KM
-8GB Gpu Target
-Epic settings,
-Ultra Wide Resolution
-RT (Propably 2 Clients, Since Disabling Hardware RT requires Shader Recompilation, it takes huge time)
GPU with DLSS 4.0 and nothing in the background which uses VRAM:)
Proof:
OBS record take to up 700MB
Had to hide Cells with Height FOG to reduce memory footprint for optimized 8k Level all Forest Runtime.
DLSS will deliver additional 400-500MB memory save
Jeezus. You do realize we were doing 144km squared 6 years ago on 4GB to 6GB VRam tops in 4k native at 60fps or better right?
You âapproveâ of something 1/9th of that overall size running more poorly than it did 6 years ago - why?
I gotta ask - are you affiliated with another engine?
Sure, because having enough common sense to notice that 2+2 equals 4 and not 24 - like Epic wants you to think - automatically means that I must be paid by or affiliated to some competition. Makes perfect sense actually. 
Iâm still trying to work out what your deal is
Iâm guessing youâre about my age (my apologies if not, wasnât meant as an insult). We kind of realize intimately that the only valuable things in life and time and energy. I wonder why youâre so invested in UE, but never seem to have anything good to say about it.
Hi, its 256 KM Squared level (16x16) , half of this in depth Loaded instantly to be visible and playable for multiplayer scenario so you can able to see whats going on in that distance - its full raytraced enviroment, 500k trees loaded into cells without any visible change with like unlimited character speed, there is 0 chunking right here. Its possible to increase visibility it more but it will be required for lower gpu to reduce details - by turning RT/Lumen off and couple seconds more to loading level.
Can You show that scenario when you fly with speed like that with the engine You mention? I feel You arent satisfied that the engine is ready to do dense open worlds with nanite tech.I took time but goal is achieved. 0 Generating Hlods,0 Baking Static things. I can turn off that RT and it will be easy 3x times more FPS and much less ram allocatedâŚ
144km squared is not 16km squared - 256km is your total area. Mine would be ~20,700 km. And there is people out there with way larger maps.
Also, you are missing a lot of things that are part of everyoneâs scenes like reflections and water - which would be the only reason for Ray Tracing to even be on.
The shadows seem to be cadcade based on the circle forming around the player, the fact the landscape has none far as I can tell.
Donât get me wrong: I am happy you are sharing your testing with us.
However at this point this âtesting" is unrealistic and it sets bad expectations from lurkers (like chatgpt) who come in and will just say to the next noob: oh yea, this one guy was able to get it working and it looks like he reports the pefromance is OK. (Worse, epic may use this as an excuse to never fix a darn thing as they have done before).
Overall, I think you have a map that is about the overall size - or less - of A Boy and His Kite. That project was trash, it had heaps wrong with it along with very poor performance - but it was far more realistic in terms of what is needed in a final scene (though no water either).
Additionally:
You shouldnt be benching the system without clouds in the sky if you will be adding them in at any point (because if you do it wrong theyâll eat all the performance out of your project). Also they cast shadows, and there are 1000 different ways to achieve that effect.
The light being dynamic should have little to no impact if you raytrace, so you can make it move faster for your benchmarks to show the shadows moving along / computing (unless you have it in cascade, then itâs pointless because all you should is glitchiness).
Now, obviously yes. On older tech things stream in and that could take time - but the player speed is managed to prevent that so it is never really an issue - at least itâs an issue you really shouldnt care about beyond the maximum speed of your player.
You can look up the benchmark for GTA5 for reference. The fastest speed is the jet, so they have a bit where the jet runs around the map. Can the old system do that? Maybe not. I think the top speed was around 60kmh before you start noticing streaming. But again, comparatively it is a fully loaded scene so itâs only normal it works a little worse.
Stuff like JustCause 4 I think used ue4 and allowed traveling faster (than 60kmh) - also streamed in stuff. Either way, I donât think itâs one of the limitations you should focus in on compared to native render size and a playable (maybe even 144 to match the hrtz) FPS.
If you do further testing and share them, please try to be fair about what you are sharing. (Not that you were purpously hiding things, but throwing a video and some keywords on a post isnât really helping either for those who will have 0 idea what they need to look out for).
@RecourseDesign epic hasnât given me anything positive to say about them in over 6 years. Generally they double down on doing the wrong thing. Everyone tells them. They usually go and do worse (and then try to pontificate how everyone is against them because of âreasons"). Itâs systemic at this point. I doubt Iâll ever find anything nice to say about them and the engine even if I go specifially out of my way to look for it.
Jeeze, I read that as 144km2 - thatâs a big map - most of the games Iâve seen where the maps are that big are flight sims, space sims or racing games where a whole other approach is needed anyway and theyâre not designed to be walked around interacting with all the environment.
GTA5 map is 76km2, quite a bit smaller. Just Cause 4 uses the Apex engine (I played that with 3d glasses for 100âs of hours - awesome game).
I agree with you about having clouds when testing, the closer you can get to full game features the better. Also, when Myasga adds more objects into the level, there will be a lot more occlusion which will change the mem/perf quite a bit. I think the focus was on how Nanite performs compared to LODs though, in line with the topic of this post.
Regarding the UE5 path - sure, Iâve had my moments when Iâve just wanted to throw in the towel out of frustration too - but I am still able to find plenty of good things about the engine, and in the back of my mind I think âI can always revert to 4.27â (one of the reasons I still support 4.27 with my products). UE5/Nanite/Lumen is more an investment for the future - weâre seeing 4000/5000 series cards perform quite well - future hardware is going to eat it all for breakfast without a stutter. It wouldnât have been feasible to just release UE5 when 75% of gamers had new 5000 series cards - it takes a huge amount of time (Iâm guessing you also have written youâre own engine by some of the things youâve written, so know how long it takes to plan, develop, but mostly - test and hone).
Not to mention that Epics baby is fortnite - supplying an engine along with support, documentation and staff is not financially beneficial, Tim has stuck to his word about giving us this engine to use, making it generic enough to create games/video production of any type - I imagine their accountants email them regularly asking them to stop. So theyâve also had to target the film industry where there is more money, so sacrifices were needed.
And donât get me started about 4K - it was released before anything had 4K - it was all smoke and mirrors, up-scaling everything - marketing bs to get people to upgrade their TVs - to compromise about targeting hardware that doesnât exist yet itâs a necessary evil to provide up-scaling in the engine. Weâre in the tween stages of the engine and hardware development still.
And finally, if a gamer is spending their time looking at the actual pixels rather than be submersed in the game - whatâs the point? I think there are people who have drawn the gamers focus away from enjoying the gameplay to hating on the pixels for their own selfish wants (not you, you know who Iâm talking about) - sad state of affairs really, itâs like not enjoying a novel because youâve been told that the font is slightly irregular.
If I wasnât able to find anything good about the engine, Iâd probably switch to 4.27 or dust off my own engine - both good reasons to look for good things in 5 
So Assemblies - Skeletal Meshes checked, Couldnt export to SM - crashed and leafs were separated,dissapear from the trunk. i Went to minimum bones to handle an tree. Comparision in performance Below.
To some extent the progress is nice, but its 4 years (rather than days) late and $1k (rather than a dollar) short.
The main issue is epic being epic. They release doctored videos claiming features they think theyâll develop eventually are already âreadyâ while in reality they havenât even been properly drafted. And people take their word for it and upgrade their projects (messing it up) to the latest version only to find absolutely nothing they claimed would work is working.
Itâs not just epic doing this btw, but they do lead the industry in terms of releasing fake features that donât yet exist (or work properly) as a way to seem relevant in a market where they really, really, really are not.
4k may have been released too early - but so has 8k. And so has 16k which is in prototype. Epic canât even do 1080p decently most of the time and ended up forcing the upscaling system (which isnât native rendering obviously) on us - specifcially to try and get to 2k/4k rendering. (With modern AI/Nvidia filters this isnât a horrible thing anymore, but it was when epic forced it. Nvidia has basically been cleaning up after them since.)
Your note on gamers and how they spend their time - it fits, but it also doesnât.
Some games drag you in in-spite of horrible things - KCD1 LOD popping for instance. Or No Manâs Sky network system (because it doesnât just have to be graphical even if that is what we are currently talking on). Iâd argue that they would lock players in and possibly be even more addictive if the issues were fixed or never existed.
The comparison to the novel should be more close to this: itâs written in Windings and you have to translate every word. Would you enjoy it? Yes, some people would enjoy it a lot. Others would not.
When something is egregious (the LOD pop in kcd1 actually) you have to learn to ignore it in order to enjoy it (just like youâd have to learn to translate windings).
So again. Iâm all for progress. But Iâm also more for âdoing things rightâ.
I donât see âoh, wait 3 years for the GFX to catch upâ to be an answer that points to anyone (at epic or anywhere else) doing things rightâŚ
Particulalry - you cant tell me that your player base could ever be expected to buy a rig with the latest xx90TI or whatever top of the line GFX name they will assign in order to experience the game as you and your team designed it to be experienced. The reason consoles are developed for and received more widely is precisely the fact that everyone sees something almost identical as the end product (which in most cases is and shouls be considered Art, so you really do not want anyone to see anything but what the Artist wanted them to see).
I appriciate you testing out the hardest possible tree to render btw. But Iâm not really too sure what it is Iâm even supposed to look at in this latest videoâŚ
Shows That Nanite cool feature need a proper documentation or shoould be fixed cause:
1mln instances spawned from Foliage Painter Pine Trees - runs at 80 FPS
1mln Skeletal Assemblies instances âNanite Foliage Beech Treesâ Runs at 6 FPS
Somehow Engine says it 7mln triangles for that assembly .
it have 29 bones so 2141 x29 give 62k⌠and couple of bones are trunk with even less in nanite visualization.
Ok, Iâm a bit confused - it looks like youâre getting 120fps in the editor with all those trees but it drops to either ~70 or ~7 when played - am I missing something?
Ignoring VR which is a whole topic - 8k and 16k are definitely going to need up-scaling - and at my age - thereâs absolutely no chance Iâm going to be able to tell the difference apart from performance.
My point about the novel was more that - even if the font is fine (as in still completely readable), people reading it will have a preconceived notion that the font is bad and will have 1/2 their brain looking for it rather than concentrating on the story. Whoâs to blame there - the printer, or the person trying to ruin the story by changing the focus from the story to the font?
Itâs like - thereâs always that one friend who speaks loudly during a movie pointing out the flaws in the story-line
I wouldnât expect them to even try to make skeletal meshes behave with nanite at those massive numbers.
The reason it wouldnât work is the same reason it wonât work currently.
First of, each mesh would have to have some kind of animation sharing to make it even remotely possible - then maybe they can translate that into an automated shader.
The second reason is the animation itself - and the LOD required for it. You can maybe get it to behave a bit better by scaling down the bone count on the LODs. But the main issue is precision - too many bones in too many skeletal meshes will always mean prohibitive calculations to apply as a converted nanite âthing".
Essentially-yes you can try to make a crowd by joining all skeletal meshes togeter into a unique asset, but to get it to actually work in engine you have to convert it to a static mesh and give it an animated material - or just go with Alembic.
Very similar here. Iâll be really surprised if they ever manage to make it happen âon its own"âŚ
with that in mind, you should probably nix the skeletal mesh idea anyway.
you always should have a static mesh impostor until the player is interacting with it - at that point you swap in the skeletal one so you can have interactions and physicsâŚ
Yes you are right. I cant answer to the question why is that. Skeletal meshes and Nanite foliage are features that should improve things. So that is not question to me why it runs so slow at playthrough with much less complicated structure of the tree 
Drop in fames come also from screen resolution that is increased in play mode.
FastGeo Allows me to skip HLOD and Impostors, At this point they take more memory and HLOD provides to crashes - to low memory to operate with them and need to have code to load them async way of course it rquire alot of work and fixing that lowpoly stuff.
for the clouds i did reduction - cause they are toooo heavy with standard preset like 10-20 FPS based on resolution.Ive reduced to like 1 FPS impact.
On the 2 last demos all light mixer components are placed.
While KJ is a loon. Your results are completely in line with standard because TSR was built for PS5 XSX AMD Radeon RDNA 2 first (utilizing the special fp16 arch) and PC second. The older cards (1080 and below) donât have the required fp16 power to properly utilize TSR with lumen,VSM, and nanite (see TSR Feedback Forum post by the person who worked on TSR, ironically KJ was still cordial with Epic staff at that time.) That being said, UE5 has made strides to increase performance which KJ always finds a way to complain and nitpick because grfting always needs a new monster of the week to rage about.
People seem to forget that when you make a game you are supposed to use the tool that best suites the process to completing that game.
