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

According to this guy: https://www.youtube.com/watch?v=dj4kNnj4FAQ

The answer is yes. Virtual textures, luman, nanite are meant to play together.

You don’t NEED to but it would be better for performance to use virtualization. As far as materials, you can still use master-materials, nanite doesn’t change that paradigm, but Substrate might when it becomes standard.

For now, you can still use your ‘regular’ material-tech on nanite with the usual caveats around WPO, etc, etc.

As far as merge, you should instead build Assemblies. This is basically merging for nanite since it can instance the entire assembly, and you can edit the assembly directly, post-creation so it seems ‘better’ on that point.

Nanite is really about topping-off on screen-cost so that triangle-count can grow and not grow the cost. Eventually, the nanite-clustering will cost you just-about a screen’s worth of detail and whatever you throw into the scene will be made to work within that constraint. It costs you more up-front, but ultimately will top-out to cost you less, much like a fixed-cost (which is insane when you think about it).

Also, nanite is for perfect-LODing, or as close as you can get. You don’t need to manually define LOD’s with the tech instead creatively-destroying the mesh as you get farther and farther away. This saves a lot of time.

I’ll be using substrate on my project. So master materials help with performance still and are necessary, got it. What are assemblies exactly? Most of my assets are coming from zbrush.

I found a neat greasemonkey script: Greasemonkey script for phpBB boards that hides posts of certain users · GitHub

@TheKJ I literally provided you with a real game benchmark using Nanite, where we reached over 180FPS using Photogrammetry Assets and 8k textures.

That was (to my Knowledge) the only valid Benchmark in this whole thread.
I also talked about why TAA is a Choice nowadays and is redundant with the use of Upscalers, as these will automatically AA Images.

Yet you still fall back to the same arguments every time this thread continues.

Multiple people have pointed out that they are working with Nanite in Production and are benefiting greatly from it.

At this Point the whole thread (like you said) is just people disagreeing with you and you defending yourself.

Why don’t we just agree, that it is just a skill issue from your Side and move on?
Because obviously your oppinion is the far far minority…

That’s amazing! Would love to know more about the project and what you all did for performance optimizations. I’ve been using all of the big features in UE5 at native 1440p and got above ~94FPS with loads of nanite foliage on screen, cars, etc… I can push over 100 FPS at native 1440p resolution with TAA set as the AA method which is on par with what I see in RDR2. I have yet to finalize my native resolution settings for TSR and expect to push it even further.

Spec: Ryzen 5950X and a 3090ti.

My mentality on all of the new systems in UE5 were completely wrong out of the gate. I didn’t realize epic cranked up everything to 300% out of the box. I knew nanite needed to be configured but I felt that there weren’t that many options available. So wrong. Found ways to configure it similar to VT’s.

I’m still working on finalizing performance optimizations for both Nanite, Lumen and TSR. There were some things that could have been improved and epic has started that by allowing control over ‘r.Nanite.MaxPixelsPerEdge’ per static mesh in 5.4.

I would love to know what you’re doing. A few other developers and myself have been discussing different tricks and findings for optimizations with nanite & lumen here.

This thread is the very definition of beating a dead horse.

How about you compile a list of these titles along with quotes from the devs specifically saying that Nanite is their issue, and up-to-date as in latest engine versions.

Without some type of proof they’re just words.

Very impressive! It really really does remind me of the roads around where I live; tons of trees, scruddy grass, not too many houses around.

I like it.

EDIT: it’s a solid-take on a base-engine. You could use it for making the Tony Goldblum’s BattleFour Field.5 game we all want. Check the Vigilante assets…

That’s funny - I really liked it for exactly the same reasons! Plus it’s got a very realistic look going on - impressive alright!

@JustinDarlington
Here take a look at this post.
(The one i replied to)

This and a few posts below are some benchmarks and tips/Tricks to improve performance.

Will take a look at the thread you mentioned later that day

Multiple things here.

you keep using the same arguments over and over again, even though i and multiple other people debunked them multiple times.
With proof.

I would even go so far to say that you did not provide any valid evidence at this point, while multiple people proved you wrong.

I am not interested in being part of your Main post. I am rather trying to stop you from spreading uninformed misinformation.

My scenario was the best benchmark you had in this thread so far. It was a full, detailed game with over 200 individual meshes and VR rendering (which is by default more demanding than normal rendering.)
It was by far a limited benchmark.

We used TAA purely as a style choice, to achieve a clean and „dreamy“ image. We could have had it removed anytime due to the use of upscalers.

Also lets be real for a second.
TAA and Upscaler issues don’t matter for the average consumer…
Thats why i don’t get why you are opposing them so fiercely…

Ask an average gamer who is not a graphics nerd or a journalist.
They will not even notice small frame inconsistencies caused by these techniques.
All they care about is a clear Image and a high framerate. Which both can achieve.

We ran multiple playtests on the gamescom with non VR Games, where we asked other players, if they noticed anything weird about the graphics. (Because we had the same oppinion as you)
The only people criticising TAA and upscalers were other developers.

The average player was saying: „The image looks very clean, like a camera recording“. (TAA)
They did not even notice upscalers.

We as developer live in a bubble. We see the problems caused by these techniques and blow them up way too much.
TAA has issues, but they don’t matter for the average consumer.
And on the contrary both of these techniques excell at their task.
TAA produces a smooth image and upscalers improve the performance drastically.

So please get your oppinions from your customers and not from other developers. Because at the end, the customers opinions are what matter.

TAA is a SOFT option.

So its optional. Glad we cleared that up

Epic has been pushing this stuff through workflow manipulation.

No they havn’t. The documentation is easy to read, and its easy to switch off TAA. Every game dev on the planet knows there are different AA algorithms with different trade offs and that they don’t have to use the one Unreal has on by default.

Zero clue to what I’m referencing.

DLSS and TAA are different things. BOTH are optional. DLSS is NOT required at all but if you are upset that AI is going to play an increasingly important role in Graphics and CompSci in general in the future, I have good news and bad news for you. The bad news is that GPUs cannot shrink in size indefinitely, nor can they grow in core counts indefinitely. AI will be increasingly important to squeeze performance out of cards. The good news is there are very smart people working on these problems for you, so relax bud. Sit back, code your game, ignore the features that you don’t like, because they don’t impact you, and use the features you do like that engineers with Phds and decades of experience are building for you.

Very interesting and amazing for VR. I’ve been building a large open world with loads of Nanite foliage and other geometry. My goal was to push some boundaries in how much foliage I could have on screen to recreate the look of my home state of GA. So definitely different in scope. Your results are impressive!

You can see in this shot what I mean by pushing that boundary with foliage and even more inside the forest. Nanite allowed me to increase the density of my foliage greatly:

image1920×1045 369 KB

this guy in the video keeps mentioning at the end of the video he will include a link of cvars… never did… never gave us that link…

Just because most people are unaware of something doesn’t make it wrong.

Popular ignorance doesn’t equal truth.

Experience doesn’t make you right, it just makes you experienced at being wrong.

Still just words - no proof - where’s the compiled list of titles in the latest versions where the devs have specifically stated their issues are Nanite?

I live in New England and this could very-easily be any road in the surrounding towns. Moreso towards the coast/Cape given the gnarlyness of the pine-tree (they get stunted from salty-air).

Obviously a game, but there’s no drop in the illusion. VERY well done.

Thank you for your kind words!
The main difference between your scene and mine is, that you have a forest vegetation which adds a lot more overdraw than my scene.

We have a landscape, a layer of grass foliage and maybe a tree/bush. But behind those are mostly rocks and hard surfaces. We specifically designed the level like this, so we can minimize overdraws with leafs.
Also our level is shaped like a bowl because it is an arena the player can play in. So there is not much draw distance, the number of meshes is limited and the mountains are acting like a hard border, not only for the player, but also for mesh culling.

Your scene has layers and layers of vegetation so it is actually really impressive on your part, that you managed to reach such high frame rates!
What helps with that, is, if you introduce hills or cliffs inside the forest. Like this:

image1920×1080 376 KB

One note on the side though, your saturation is way too strong. Your image looks too green, to the point of it being Unreal (pun intended ^^)

https://dev.epicgames.com/community/learning/knowledge-base/9Jx9/unreal-engine-optimizing-ue5-advanced-rendering-graphics-performance-and-memory-management-resources

They just forgot the link in the video description…

That’s awesome! Adding more mountains/cliffs to remove some overdraw is something I’ve been looking into. That’s definitely a pro tip [GG]. It’s finding that balance to capture your original idea plus performance that’s really challenging. You’ve done this extremely well it seems! Your scene is stunning!

For me, this boiled down to configuring nanite and lumen systems through CVAR’s. I also ended up altering some elements of the engine to allow me to utilize nanite more efficiently with RVT’s and landscape splines. And then optimizing any materials that were taking a long time in the nanite rasterizer.

GOOD! I DID MY JOB LOL! I originally matched IRL colors but found it boring. I just find general realism boring period. I love the more stylized look of the colors. But I do understand what you’re saying and hope to clean it up more. A clean stylized/realism is what I’m aiming for