Preach! Your results are amazing and my project is also close to that level of performance (60fps on a GTX 960).
The guy doesn’t understand the scalability of actions: the tests he’s showing are just a bunch of assets, while a scene has hundreds of them. Nanite helps with instancing, culling, and skips the whole LOD creation and setup. While also giving much higher fidelity and lower impact on memory. It’s a no-brainer… if you ACTUALLY make games.
4 Likes
Nanite was arguably bad in 5.0 but I believe it was 5.3 or 5.4 they gutted all the legacy patchwork stuff from UE4 with it and it got a lot better. My game runs with everything being nanite including characters at above 60fps on a base ps5 and above 90fps on my PC, I would get like 10-20fps without it.
4 Likes
You kinda seem like you know what you are doing… So then, why do you go on a performance thread and give halfhassed info?
at what resolution do you get 60fps?
I take that back. Scrolled up a few posts. You obviously have 0 idea what you are doing, its not even worth pointing out why.
1 Like
lol okay, keep trusting the con man dude.
3 Likes
Speculating here, but thats likely just the rendering thread, in dx12 it probably does operations it cannot conduct in dx11. Think of it more simple, like having athmospheric fog on or off.
If you want more precise, then you should learn how to open up abrender benchmark and compare the 2… it does tell you the timings.
I’ve tested the performance between dx12 and dx11. Depending on the GPU, it can have a greater or lesser impact. For example, a 1050ti is more affected by dx12 than dx11. There’s no difference on the 1080ti. Both graphics cards support dx12.1
Because Nanite has an overhead, but once you get above that, it’s highly scalable. Dallas Drapeau explains it here: https://www.youtube.com/watch?v=Vzz8_O3PIUg
So, if you just do some rando tests with a handful of elements and polygons, it would behave badly. But if you start throwing at it literally millions (if not billions) of polygons, that’s where Nanite really shines.
1 Like
So the documentation is right, for low polygon it gives bad performance but someone from epic said that it worked for both high and low.
From the documentation :
What Types of Meshes Should Nanite Be Used For?
Nanite should generally be enabled wherever possible. Any Static Mesh that has it enabled will typically render faster, and take up less memory and disk space.
More specifically, a mesh is an especially good candidate for Nanite if it:
Contains many triangles, or has triangles that will be very small on screen
Has many instances in the scene
Acts as a major occluder of other Nanite geometry
Casts shadows using Virtual Shadow Maps
So the documentation is very clear to not use it if using only low poly geometry and without many instances (it’s good for a Lego/Minecraft-like game too with thousands of same instances)
1 Like
Because Nanite has an overhead, but once you get above that
yeah, base your logic on a UE5 dev meme that got spread around due to ignorance o the topic:
This admittedly poor description of Nanite’s performance is 100% wrong. The act of enabling Nanite only cost around .54ms at 1080p on a Desktop 3060. Mesh shaders are extremely inefficient & SW rasterizer is slow and scales with complexity because of how inefficient cluster culling is. None of your sources references properly prepassed content performance either. The content you shared used butchered content/rendering approaches for non-nanite to premote performance on 4080 hardware. You gave no side by side comparisons of mesh shader & SW raster performance for foliage either in any of your post.
SW raster grass will perform worse than HW shaded masked materials because my team proved how much faster mesh shaders are.
Nanite should generally be enabled wherever possible. Any Static Mesh that has it enabled will typically render faster, and take up less memory and disk space.
So the documentation is very clear to not use it if using only low poly geometry
No, it will not “typically” run faster & official presenters are saying the opposite of your conclusion: https://youtu.be/Cb63bHkWkwk?t=5496
It’s also manipulating for the docs to say a prepass is 75% faster than a basepass to jusifiy Nanite rendering. Epic should include non-vrs shading rates based on quad rates for prepass & non-prepassed content vs the visibility material resolves & vis buffer raster methods.
My team provided this for 5.6, but it should be standard for each version on the docs.
Epic should also fix the viewmodes for Nanite for anyone crazy enough to use it.
Nanite is a constant bottleneck for games, the documentation is a part of that problem but more so the people who think this topic is based off “opinions”. The facts have been provided, people
You have been proven wrong dozens of times (here, on several subReddit and on Youtube) by dozens of different people who have conducted real tests, unlike yours. When will you stop being hypocritical ?
Nanite is using hardware raster for big triangles and software raster for small triangles because each of them are faster respectively in these conditions.
We don’t care how much time takes pre-pass only it’s a whole. You never show the total frame time and that’s what matter.
2 Likes
You have been proven wrong dozens of times (here, on several subReddit and on Youtube)
Hasn’t happened once. They’ve always botched traditional rendering & people have called out for doing test that don’t even perform well & not taking the traditionally raster path seriously in terms of optimization.
It’s just a bunch of people say “look! my 14 billion triangle scene performs better with nanite, dErBunKED!” and pretending this is the conversation while their scene performs worse than several better looking games (and their geometry timings)
We don’t care how much time takes pre-pass only it’s a whole. You never show the total frame time and that’s what matter.
I rest my case Epic Games.
You’re misunderstanding what mesh shaders are and why they exist. They’re not inefficient… you’re just approaching them with the wrong expectations. The traditional graphics pipeline was rigid and built around fixed stages like vertex and geometry shaders, with the API handling how data flowed through them. That old system imposed limits on what you could control or optimize.
Mesh shaders throw that entire structure out. They give graphics programmers direct control over the pipeline which opens the door for even better GPU performance.
“you’re just too stupid to realize Nanite is good” “Nanite is not for beginners”.
Got anymore gaslighting statments?
Pretty words with NO meaning, just bullcrap marketing.
Alan Wake 2, Monster Hunter Wilds & every last UE game that uses Nanite including fortnite are proving Mesh Shaders as an incompetent approaches to geometry rendering (that’s 3 different engines & programming teams)
Mesh shaders are nothing in comparison with SW raster where youtubers are falsely claiming it’s “3x faster”. Even Brian Karis mention that in the context of unoptimized prepassed content.
You’re misunderstanding what mesh shaders are and why they exist. They’re not inefficient… you’re just approaching them with the wrong expectations. The traditional graphics pipeline was rigid and built around fixed stages like vertex and geometry shaders, with the API handling how data flowed through them. That old system imposed limits on what you could control or optimize.
Mesh shaders throw that entire structure out. They give graphics programmers direct control over the pipeline which opens the door for even better GPU performance.
At this point, if you’re not willing to sit down and write an open source DX12 application yourself to learn these basic concepts and showcase your progress of knowledge, just drop it.
I also noticed that you took some of my points from my earlier post and used them to mislead your audience on YouTube a while back. I would never agree that mesh shading is the problem however I have stated and will state again that mesh shaders opened up a huge learning curve for graphics programmers that never wrote graphics code on such a low level before which has caused the industry to slowly move towards proper implementation.
4 Likes
You’re beyond help. There’s no point discussing anything with you. You think you’re better than everyone you haven’t shown anything conclusive since the very beginning of this topic.
You’re completely closed off to discussion and reject any ideas other than your own.
Can a moderator lock this thread ?
This is not a discussion thread just someone biased opinion.
5 Likes
Yep from 5.4 and DLSS4 - there are massive improvement in RT so no matter TheKJ says he is not right there 
He just cant accept that. Before that i was LOD guy too but things changed when done Open World Enviroment tests. Then checked memory allocation on GPUs and massive difference with distant visuals. Also Shadow Maps Flicker with dynamic Sun Positions even with LODS - Virtual Shadow Maps does not, DLSS 4 wont help in that case.
Did the creator of that diagram give you permission to use it?
5 Likes
Good question because he answered me that no one proved him wrong since he copyright strike every videos talking about him and delete every comment proving him wrong.
You can check his videos curiously none of them have comments saying he is wrong or negative comments.
4 Likes
I’m honestly curious what he’ll do next. The topic is pretty niche, and I doubt repeating the same content will keep people engaged.
2 Likes
Not necessarily, a low poly scene may benefit from it if they use a lot of grass, trees, and other meshes or have a lot of characters like a total war type game.
Nanite has 1 initial hit and then almost 0 difference no matter how much you add to your level post that initial hit, if you are just flipping the third person template you won’t see a need for it since it’s 10 cubes in a tiny map, but if you took an actual production scene like a fab asset showcase map and turned nanite on or off you would see a difference.
Game devs don’t like to skimp out and cut scope for level design, look at the difference in environment detail between AC Shadows(not talking about game quality), and AC odyssey or Valhalla, shadows has way more density to asset use than previous games because they use virtualized geometry. Looking at GTA 6 trailers they likely use virtualized geometry too.
1 Like