Nanite Performance is Not Better than Overdraw Focused LODs [TEST RESULTS]. Epic's Documentation is Dangering Optimization.

Hey @TheKJ , i see your point but as i shown there is massive difference with thickness in display with Nanites. So in that case there must be consideration if LOD Foliage assets can achieve that in UE5. If not there is needed Rank of that visual quality of that in % score improvement - as example Vote from Devs,Players and transpose that to the frames lost caused by that quality increase.
So if the Masked LOD foliage assets can be displayed 1:1 as Nanite with also better frames i shown - please create some map.
Give us tutorial - how to create that Meshes,materials for them and how to setup engine.
In final how to achieve faster game render - and i will forget that Nanite exist :slight_smile:

Today iam doing deep test on my 16K terrain and nanite slows down here in every scenario packing stuff with Nanite Foliage 20k instances in 2tiles (1024 terrain tiles are placed on the level)


Setup with only Directional Light

Standard LOD+HLOD Terrain score


Loaded both tiles with foliage

Full Nanite Terrain Score (Nanite Tiles and Nanite HLOD)



I thought the more the triangles packed into Nanite i will get more improvement - but it fails with in my case - Static Mesh Tiles scenario.
Going also Test for Landscape system and how things goes (Max 8k size possible here for my PC).

Thanks in advance!

(post deleted by author)

1 Like

@TheKJ This whole post just seems to exist to prove that Nanite performs worse than LODs… Why are you so OBSESSED with explaining that Nanite performs worse than classic LODs? WHAT?! PRECOMPUTED LODs are FASTER than REAL-TIME COMPUTED LODs?! NO WAY?! Who would’ve thought…

Thank you for telling me something every normal human with 3 brain cells already knows. What I don’t know, however, is: WHEN can I use Nanite, WHY should I use Nanite, and WHEN should I not use it – and WHY not? #UseCases

In your eyes, does Nanite have no strengths, only weaknesses? Why is EVERYTHING negative…
It’s not like you’re entirely wrong with what you’re saying, but in my eyes your posts offer absolutely zero added value. The way you present all this isn’t scientific or serious (at least not HERE in the Epic Games UE forum) – it reminds me more of a 16-year-old emotional kid screaming because something doesn’t go their way.

But that way you’re not going to convince anyone to stop using Nanite or whatever your intention is. At most, you’re going to rile up the gamer community against developers and/or Epic Games and stir up hate instead of uniting communities constructively, finding compromise, or even driving change in the industry.

I don’t think this is the right approach – and that’s a real shame, because I do watch your videos from time to time and you do come across as someone who knows what they’re talking about and wants to make a difference.
But with this approach, you’ll achieve the exact opposite… wasted potential, really sad.

1 Like

No one needs to explain it, explains itself. Followed by Lumen and Virtual Shadow Maps.

If it’s self-explanatory, then there’s no need to keep this post either.

I wonder when this hilarious discussion ends. Once in a while when I am bored I come here to be amused.

6 Likes

The documentation needs to be fixed, as the creator of the thread, my issues has not been addressed. The documentation is wrong & even Epic speakers are giving worse information than my studios videos.

The thread is here for engine updates & keeping track of visbuffer & nanite basepass efficiency against properly optimized content.

No point in locking the thread, more people will make Nanite performance thread but the mods need to remove rude & off-topic comments.

Dude, there’s not a single one of your tests that’s been done correctly. There’s not a single time you’ve packaged and used the recommended tools and not a single time you made a realistic test scene that would look like a plausible game use case.
There are people having double the performances in packaged build vs in-editor so all timings could be really different when packaged (UE5.6 is slower than UE5.5 in editor but faster when packaged).
And everybody knows that’s there is no “studio” we all know that you are on your own.

4 Likes

Same :slight_smile: Just lurking. Still funny coming back once a month ^^

2 Likes

What’s even funnier is Kevin refers to himself as “our founder” in his twitter posts

4 Likes

Kevin is what you get when you platform people more concerned with being youtube famous than presenting anything of substance. Every video, every post ridiculed, then he turns around and says the industry is getting better because of him and him alone XD

This image represents Kevin perfectly

3 Likes

He is the pirate software of “graphics”

oh hey @dallasdrapeau!
Just watched one of your videos, great to see you mate…

Zero responses about the performance or data given on this thread because people refuse to acknowledge the performance & visual issues with Nanite:


Mods should do their job and remove off-topic (rule breaking on personal harassments) comments.

There are people having double the performances in packaged build vs in-editor so all timings could be really different when packaged (UE5.6 is slower than UE5.5 in editor but faster when packaged).

I’ve shared several packaged test. This also doesn’t affect the shader timings. It just removes some editor timings and reduces a few memory issues(bottle necks if present, which non where). You’re also not proving major improvements relate to Nanite because you all refuse to use proper optimization tools.

test scene that would look like a plausible game use case.

No, instead I’ve shown non-nanite games rendering pipelines running faster & proved how impossible it is for nanite to be faster than optimized rendering.

Unlike all the commentators making personal attacks, I made the thread so that Epic would fix the docs which ridiculously states a typical non-nanite scenario has a 25% faster prepass than basepass.

Consumers & devs demand changes to the docs that are less manipulative in nanite’s favor.
Prove me wrong or get out of the thread & stop embarrassing yourselves. Provide internal res/pixel count all pass timings, hardware, quad overdraw, velocity settings, and then show it looking better in motion/no subpixel aliasing without TAA.

PRECOMPUTED LODs are FASTER than REAL-TIME COMPUTED LODs?! NO WAY?! Who would’ve thought…

Apparently Epic Game teachers & the docs themselves. Several people on this thread kept lying and saying Epic never said “use it on low poly” or “it’s faster” meanwhile several presentations said otherwise. Thank you for reminding everyone how important this thread was.

1 Like

Well, it’s true that it improves performance in cases where everything is overloaded. But it has a cost for optimization.
I would also like to know why the hell DX12 has more cost than DX11.

This whole thread should be deleted, there is no informations of value anywhere.
OP decided that Epic Games was wrong no matter what and that only he holds the truth. If you’re looking for impartial tests, this isn’t the right place.
The first sentence of the thread by him “There is no need for mindless scrolling over opinions and arguments for the facts” => so essentially : “LISTEN TO ME EVERYONE, ONLY I HOLD THE TRUTH AND I AM BETTER THAN EVERYONE ELSE”

CD Project Red the second biggest studio after RockStar Games that have many engine programmers switched to UE5 while they already developed a full game engine themselves. They managed to make a huge pine tree forest with every single needles rendered in 3D literally the worst game scenario you could possibly imagine and used Nanite, Lumen hardware RT and VSM and made it run on a base PS5 at 1080p 60 fps with zero pop-in using a new voxel based rendering for vegetation that very recently have been made available on UE 5.7 main branch. Lumen seems to have been greatly improved too because we can see a major change in lighting (going from night to daylight near instantly) and lumen managed to update very fast wich is currently one of the biggest lumen issue so they seems to have managed to fix it too in this realtime demo.
Even Ubisoft made a Nanite-like system for Assassin’s Creed Shadows (they named it “micropolygon geometry system”)
It must be that every major engine and game developers are stupid and you outsmart all of them right ?

4 Likes

the test i did shown that nanite assets are denser in the scene and last LOD with imposter took more memory usage for instances. For Open World there is nono to back to LOD. You will get some improvement in Frames but you wont be able to pack that much assets for 8GB Gpu-s with having “similiar” quality in a realistic type of project.

I took another look at the visuals of Death Stranding 2 in different scenarios, which is considered to be one of the most optimised and visual impressive game to date and there’s a massive drop in the visual quality of mountains at middle to far distances, especially on the base PS5 running at performance 60 fps mod.

DS2 Screenshot

DS2 Forest

In the meanwhile The Witcher 4 demo is so much more impressive with zero visual degradation on the distance and the use of raytracing for lighting and shadows. The forest was so much more detailed it’s not even comparable (DS2 forests have flat terrains with very few grass patches and small rocks even on PS5 Pro quality setting). The city was running good with a lot of characters on screen. We will have to see how it looks on the final game in 2-3 years but it was a realtime demo presented on stage with much more optimisations to come.

TW4 Screenshot

TW4 Forest

Looking at the screenshots, there’s no way they could have done it with classic LODs.

1 Like

Hopefully you don’t delete my comment again, Kevin. You are not properly comparing Nanite to traditional rendering, as the other 150 people in this thread have rightfully pointed out to you. Yet you keep lying about Nanite being a bad rendering option when you keep purposefully omitting its purpose and making disingenuous comparisons. If all you have is 1 mesh, a skybox, and some cubes like some of your YouTube videos do then Nanite is not for you. It’s built for games like Hellblade 2 that have extremely dense level detail that traditional rendering wouldn’t work and would have extremely low performance. Indies can use this too to make bigger and more dense worlds with more assets without worrying about making custom LODs and shaders for everything.

It’s really annoying how you censor everyone who criticizes you while complaining when you get censored. Why should Epic Games defend you when you are their number 1 griefer? Your entire life is devoted to hating on them and taking credit when actual developers make advancements and contributions.

And just for the record, until I get a message from a moderator explaining why I’m being censored, I’ll keep correcting you if this gets deleted as I really don’t like deceivers.

2 Likes

At this point, I think it’s a running joke for the Epic staff to allow his crash out over nothing. I’ve been able get Nanite and Lumen running smoothly on the Xbox Series X and S at 60FPS (Performance) and locked 30FPS (Fidelity)

Screenshots captured directly from the Series X running at native 4K with cinematic scalability(I modified the DefaultScalability.ini to reduce the default cinematic quality of course).


I was initially frustrated with the state and lack of proper documentation for Nanite and Lumen, but after playing with it and ACTUALLY BUILDING A GAME with these technologies, you learn a whole lot. Also, the newer version of the engine have brought forth great improvements to all of the technologies despite what KJ wants to say.

You will never understand how to use these technologies properly unless you build a game with them. KJ has been building a youtube channel instead so of course, there is no real knowledge.