Nanite, Lumen, TAA. Can they theoretically become human?

[Macro Device]

I mean, these things ruin both visuals and performance in recent games. I am not a game development specialist, that's why I'm curious if that's because all three features are doomed to fail or it's because we're yet to see a studio that use it correctly.

On paper, this should massively improve the development speed. Also, on paper, this makes things cheaper and easier for developers. However, recent titles make it clear that games from 10 years ago looked better at 1080p than newest titles do at 4K, minus the enhanced textures and (likely) normal/depth maps. Not to mention GTX 970 ($350ish on launch, under $50 today) runs these 10 y.o. games smoother than $1000+ devices of today run recent games.

Can this trio enhance our gaming experience? Or no matter how you try, the end result is doomed to be full of ghosting, smearing, and flickering artifacts?

 

[Vayra86]

The implementation is just lacking and probably needs a lot more work to be great.

The overall reluctance of Epic to actually get serious on this point is staggering though. There is also an industry push to short time to market, and they think this is the way - and sure you can poop out a lot of product this way, just don't ask how it plays.

 

[Onasi]

You’ve piled quite a bit into one. Nanite and Lumen are, theoretically anyway, meant to simplify things for the artists. Whether or not the way they are implemented actually is worth the trade offs in the end for GAMES is another question entirely. But, I suppose, seeing how Epic nowadays really pushes the whole “look, our engine is used for CGI in big movies” they are seemingly not caring. Certainly, I’ve yet to see any UE5 game that would show that developers (apart from Epic themselves, natch) actually have a firm grasp on the tech.

TAA is just a crutch that exists because every other AA method can’t really work well with modern rendering techniques. It’s not really meant to “enhance” anything, just be an “acceptable” AA for current times. Problem with any temporal algorithm is that it actually wants higher resolution and framerate to work better, but the games are getting heavier and heavier. It, or at least its variants, can be potentially tuned well - there are DLAA implementations that are rather visually stable. Even TAA itself is often misused - I distinctly remember Scarlet Nexus having dogshit default UE4 TAA that was easily fixable with, like, couple of line changes in .ini for no performance loss.

 

[Macro Device]

every other AA method can’t really work well with modern rendering techniques

Is it so? Why?

Problem with any temporal algorithm is that it actually wants higher resolution and framerate to work better

I still see a lot of crap going on even at 200+ FPS which is ridiculously fast if you ask me.

default UE4 TAA that was easily fixable with, like, couple of line changes in .ini for no performance loss.

I was googling the same but for CP2077 but never achieved anything. Do you have a clue how one edits TAA in this one? Unfortunately, disabling TAA altogether is only worth it at 4K+ and with upscaling being removed from the equation (doesn't work sans TAA), performance on 6700 XT is in the gutter. So I'm curious as to what can be done to improve it rather than make it a riddance.

 

[Onasi]

Is it so? Why?

Multi-sampling doesn’t really work with deferred rendering and isn’t effective for transparencies anyway, which modern games are full of. Post-processing methods (your FXAA/MLAA/SMAA) suffer from heavy full-frame blur (though this can be mitigated) and often fully shit the bed in motion. This is solved by adding a temporal component (like SMAA T2x) and at that point we are full circle in “TAA bad” territory. So devs just choose the path of least resistance and use TAA by default, seeing as how modern engines already include it and upscalers require temporal components anyway. Hell, often games aren’t really meant to be seen without TAA since developers started relying on it to smooth out dithering, for example. As an experiment, try running, say, RE2 Remake without TAA and look at characters hair. It’s a mess.

I was googling the same but for CP2077 but never achieved anything. Do you have a clue how one edits TAA in this one?

Haven’t touched CP2077 since release, so can’t help there. Obviously, it won’t use usual UE files and tweaks. Can try FSR Native AA, heard that was in the game? But if your concern is performance, well, you’d have to use upscaling. TAA is itself lightweight comparatively and the example I gave above with SN was to make it look better (less shimmering and ghosting in motion), not to increase performance.

 

[Macro Device]

As an experiment, try running, say, RE2 Remake without TAA and look at characters hair. It’s a mess.

No, I know how bad it is if the game designed around TAA runs with it disabled. Not my point.

Can try FSR Native AA

It achieves better quality than rendering from lower resolutions but doesn't fix one of motion ghosting issues introduced around March '23 or so when they released the version 1.62 and never attempted to fix FSR. I got Optiscaler to utilise non-vanilla FSR 2.2.1 kinda reasonably well but it still has some misses.
My thought process was like "Maybe it's not only FSR but also TAA that needs tweaking" but I only know how to disable it and it only works well at resolutions you need something faster than RTX 4090 for.

I'm also still waiting for Intel to fix the flickering issue in XeSS. I'm totally happy with how XeSS handles everything else but man, these flickers are blinding.

 

[Onasi]

No, I know how bad it is if the game designed around TAA runs with it disabled. Not my point.

I wasn’t even making one. Just explaining why I doubt we’ll be rid of TAA as a default option any time soon. Not that I like it either, I find most modern games to be incredibly soft and blurry, to the point where the strive for realism seems pointless. I far prefer perhaps technically “inferior” graphics of olden days that were sharp, readable and often well-styled to be rather timeless. If we go by the latest release, I vastly visually prefer the OG SH2 and ESPECIALLY SH3 to the new Remake. And the incredibly soft and artificial look is definitely the reason why, for which TAA is responsible in many ways.

Rolling it back a bit, Nanite and Lumin are still in a state where I’ve yet to see the game that actually uses them to produce something actually impressive. I feel like where we are at the point in graphics development where I can say that modern high-budget games look “technologically” so similar that it all blends together. And I don’t see how making devs rely on these tools helps this issue.

Anyway, fuck 3D anyway, embrace the classics and go back to hand-drawn sprites. Something like Unicorn Overlord shits visually on any AAA slop.

 

[cvaldes]

Like all new graphics technology improvements, it takes skill to implement these new techniques correctly.

For sure typical TAA implementations are inferior to other anti-aliasing techniques. There are various online tutorials and analyses that compare anti-aliasing and/or upscaling techniques and often DLSS comes up on top, especially when looking at full-motion rendered footage (not still frames). This is probably why TAA isn't being pushed as the Next Big Thing in 2024.

As for Nanite and Lumen, the simplest thing to do is to look at Fortnite. Yes, Epic is using these technologies in this title. Both can improve imagery but it is very important to stress that what it is doing is making graphical improvements on the same hardware. However from a practical perspective, they are likely using it to bring better quality graphics to lower specced hardware: like Nintendo Switch. It's not really meant to make things look better for people rocking RTX 4090s. It's for people on modest hardware (handhelds, notebook PCs, smartphones, etc.) to enjoy better visuals because there are more users on those devices than flagship PC GPUs.

We have seen this with each new game engine generation. There are programmers who learn how to harness new technologies and there are programmers who fail at achieving that mastery. Unfortunately the latter greatly outnumber the former.

In any case in 2-3 years game programming will get a boost from AI assisted coding. Or at least the studios that use AI judiciously will likely have better results than those who stick with doing everything the old fashioned way. But AI-assisted game development won't be an autopilot thing. It will still take some level of human guidance. And AI can't write good stories yet. But it should be able to speed up much of the development of the technical side of things.

 

[Macro Device]

fuck 3D anyway

Technically, it's our only option from the physics laws standpoint. All my exes were 3D at least...

So far the only truly remarkable thing about modern AAA games is that they still, somehow, sell well.

DLSS comes up on top

It's built of TAA. It's just DLSS has much more refinement than "stripped" TAA does.

to enjoy better visuals because there are more users on those devices than flagship PC GPUs.

Older methods work MUCH faster and produce more visual quality than these two. Fortnite is graphically impressive by... 2004 standards perhaps? However, no way you run it on hardware from 00s.

 

[cvaldes]

Well, you asked about Nanite, Lumen and TAA (which I consider to be a legacy anti-aliasing technique). In any case, I don't think 3D graphics programming routines can become human.

Remember that ALL computer graphics are FAKE. Every single method. Blinn, Phong, Gouraud shading? Polygonal models? It's ALL fake.

If you miss the good ol' days of Tomb Raider I, Doom, and Super Mario 64, that's fine. I get it. But remember that those games aren't any more "real" than the latest and greatest patch of Elden Ring, Fortnite, or Stellar Blade.

Now you have turned this into a screed which you are free to do. It's your thread.

At this point I hereby bow out of this discussion.

 

[Macro Device]

We have seen this with each new game engine generation.

Has never been that bad.
TAA-based AAA titles are the first in the history to be graphically worse than predecessors despite taking up more than five times the calculating power.

Now you have turned this into a screed

It might look like that but it's not a screed. I genuinely am interested in where things have gone wrong and I don't still have a clear answer to that. The fact I quote-unquote need a 700+ dollar GPU and upscaling or lowering settings drastically to run games at 1080p60 is impossible to ignore.

 

[Klemc]

Yesn there is a "worst effect" in newest/latest games bc of this blur globally it dosn't matchs the intention, so it looks false.

 

[Shihab]

minus the enhanced textures and (likely) normal/depth maps.

Has depth maps really progressed at all in the last decade (or since inception)? Depth maps are usually generated by the API (d3d, opengl, vulkan) at runtime, not the artists. I don't see why a 2024 would need to have one different from, say, Half-Life 2.

On the main topic, "better graphics" is a tricky -and highly subjective- thing to define, especially since (a) most people can't tell what's photorealistic from what's flashy, and (b) it's difficult to separate aesthetic quality from rendering one (assuming you can measure either).

That said, personally, I don't think 2014 games are better looking or even on par with 2019 games, leave 2024 ones. Things may have slowed down in the last few years, but that's the sad reality of render gains: it costs exponentially more to increase any given quality metric.

I'll give you one thing tho, the 2010s was an exceptional decade for games. Even without the inherent challenges of CGI progress, it would've been difficult to top that decade.

 

[Macro Device]

Has depth maps really progressed at all in the last decade (or since inception)?

Not a single clue, I'm just speculating at this point.

I don't think 2014 games are better looking

At least they don't exhibit this much artifacting, ghosting, smearing and insufferable pseudo-aliasing at 1080p. Of course the textures are much poorer quality but that's the only thing I see significantly improved last decade if we don't account for RT which is still in the unobtanium phase because what we see in modern games is about 2 percent of what full ray tracing is and even 4090 struggles to work with it. Maybe they come up with better RT implementations but I can't be sure on this one.

 

[yfn_ratchet]

When it comes to newer tech like Nanite and Lumen, it's a classic DLSS tale. It will absolutely wet the bed, as any underdeveloped toddler will, on the first pass and miss widespread adoption, instead needing to make the iterative improvements that should have been going on in the background pre-release until it's passably good so it can claw its way up the charts. TAA is just an 'it sucks, but it's an all-rounder, easy to implement, and helps smooth over graphical challenges we don't have the time/resources to fix' kind of thing. It will probably get superseded by upscaler-based AA.

As spaghettified as Source/Source 2 is, it's a pretty solid engine feature-wise and exemplifies what can be done with a competent foundation and a well-paced dev team. Half-Life: Alyx is an absolutely stunning title to just... look at, nevermind how mind-blowing Half-Life 2 was while still performing great on low-end machines of the day. I would even hazard to mention Respawn's spin on the engine that's STILL powering Apex. I think that game looks pretty good.

The answer to performance challenges seems to always circle back to intelligent development, and unfortunately the largest and most ambitious projects are severely lacking in that due to sprawl and too-many-cooks syndrome at multiple levels. Technology is not a glue-on patch, it's a square of cloth. You can't slap it on top of any issue, ESPECIALLY not in layers, and cross your fingers.

 

[Macro Device]

Half-Life: Alyx

Doesn't look bad indeed. Might as well get it and play at some point.

Technology is not a glue-on patch, it's a square of cloth. You can't slap it on top of any issue, ESPECIALLY not in layers, and cross your fingers.

I agree. But are Lumen and Nanite complete garbage or one doesn't need to rewrite them from scratch but rather make devs more interested in actual optimisations thereof?

 

[yfn_ratchet]

Doesn't look bad indeed. Might as well get it and play at some point.

I agree. But are Lumen and Nanite complete garbage or one doesn't need to rewrite them from scratch but rather make devs more interested in actual optimisations thereof?

There is potential in both technologies, but as it stand both are too lacking and too slow to be viable as a replacement for the rasterization tricks and busywork of old; that shows in how often new UE5-based releases get horrific audience reception for, obviously, performing so poorly that the game can't be enjoyed except by people with multi-thousand dollar systems. They need more time in the oven, as promising as the ideas are.

As is, I think the technologies are best implemented in part or very selectively to exemplify what makes them better than the old stuff. Character/viewmodel lighting, dynamic lights e.g. flashlights, lanterns, fire, explosions (Lumen), and middle to far distance objects (Nanite).

Automatic, high-granularity LOD is an especially wonderful concept, especially if it can solve a lot of pop-in issues with far-off elements that can take you out of the experience while providing a minimal or even positive impact on performance. The limiting factor, were I to spitball, is the CPU time cost, but I've a feeling we can borrow from certain voxel game optimizations by introducing a loading/computing method that spans multiple game ticks instead of one to mitigate frametime spikes.

 

[Macro Device]

the CPU time cost

I see that as an essentially no-problem because 300 dollars buy you a ridiculously fast last gen CPU ez pz. Used 13700s go for this money where I live at least and it's by no mean weak processors. Most games are heavily GPU limited if this is your CPU, some even at 1080p. GPUs are far worse if we speak dollars.

if it can solve a lot of pop-in issues

Yeah, no one objects against that. That's for sure.

So we need the features polished instead of vanished, understood.

 

[Shihab]

At least they don't exhibit this much artifacting, ghosting, smearing and insufferable pseudo-aliasing at 1080p.

Are you referring to upscaling/interpolation/other temporal filters' side effects?

RT which is still in the unobtanium phase because what we see in modern games is about 2 percent of what full ray tracing is and even 4090 struggles to work with it

I don't follow RT performance that much, tbh. From what little I recall (Alan Wake 2, Far Cry 6), even a 4070 can get by. I believe some implementations are forgiving even for older, mid-range cards.

nevermind how mind-blowing Half-Life 2 was while still performing great on low-end machines of the day

Half-Life 2 also slowed to crawl once players start flipping those toggles on, even on "mid range" hardware (although I'd hardly call a $700 CPU "mid range").
The industry's typical frametime targets does seem to have increased between then an now. Blame it on seventh gen consoles and the shift from CRTs to LCDs. But other practices -largely- remain the same: it's the user who ultimately gets to determine the actual performance by adjusting the renderer settings. While back in the 2000s people would happily lower resolutions, sacrifice AA, and elevate costly configs to legendary status, today many players are entitled brats exemplifying the Dunning-Kruger effect chaperoned by even bigger morons given podiums by youtube, tiktok and whatnot.

 

[Calmmo]

Most of these features from UE5 are barely ever used, can't see how they could "ruin" games when they go mostly ignored. Also, Epics TAU or w/e they call their latest version TAA is great as long as it's implemented well and not asked to upscale some stupid low res like 720p to 4k (not to mention some teams chose to go with the far inferior FSR instead for some bizzare reason). Look at the recent release of Metaphor Refantazio that has no TAA for example, everyone (and rightly so) went nuts over the crazy amount of aliasing in that game.

I do find it funny tho how epics marketing has brainwashed everyone so much that they think these features specifically are either a savior or a problem depending on which way the wind blows that day. Lumen brings decent cheap RT lighting, VSR will hopefully replace shadowmaps on most projects and nanite can eliminate lod pop in. Problems with all these features stem from poor implementation, not the features them selves. Like suboptimal denoising, or with nanite just blank slate implementation wihtout properly dealing with quad overaw and putting in the extra work to do topology optimization.

Heck the actuall problems with with UE have mostly to do with asset loading stutters, shader comp and poor CPU thread scheduling , things that again can be at the very least partially "fixed" by having a skilled tech team, but realistically epic them selves should make it so that the engine works better out of the box since clearly this doesnt apply for most teams/projects.

 

[londiste]

Rolling it back a bit, Nanite and Lumin are still in a state where I’ve yet to see the game that actually uses them to produce something actually impressive. I feel like where we are at the point in graphics development where I can say that modern high-budget games look “technologically” so similar that it all blends together. And I don’t see how making devs rely on these tools helps this issue.

Using Nanite and Lumen with high settings like the techdemos do, comes with a very sizable performance hit.

A large part of benefit from both Nanite and Lumen do is not necessarily visible to us as consumers. Both make it much easier for developers to create a pretty game. A lot of hardcrafting LOD levels and handplaced light situations tend to fall away. Well, the general quality of Unreal Engine helps of course. An example - look at Robocop Rogue City. This is a very AA game and developer which honestly does not show in the end result - at least when visuals are concerned.

RoboCop: Rogue City on Steam

Become the legendary part man, part machine, all cop hero and deliver justice in Old Detroit.

store.steampowered.com

Digital Foundry had a nice story about 1st generation of UE5 games. Interesting to note how games tend to use Nanite or Lumen and very few use both. And especially for Lumen what games have is usually a subset of its possible settings:

Brilliant visuals and growing pains: examining the first generation of Unreal Engine 5 games

Unreal Engine 5 games have a reputation for pushing graphical boundaries - with performance issues and other woes. Alex Battaglia grades the 1st generation...

www.eurogamer.net

 

[Vayra86]

It might look like that but it's not a screed. I genuinely am interested in where things have gone wrong and I don't still have a clear answer to that. The fact I quote-unquote need a 700+ dollar GPU and upscaling or lowering settings drastically to run games at 1080p60 is impossible to ignore.

Stop looking at the technology. Start looking at the market.

Gaming slowly outgrew itself from being a form of art (the art of the coding itself, custom code to even run games/express concepts) to being a more managed form of art (off the shelf engines and tech) where creative expression would drive graphics (good drawings make for good sprites)... to becoming something like the trio of KFC, Burger King and McDonalds.

Today, you pick an engine to drive-through, select the menu, pay the royalties and eat through workflow-code. Once you've seen the menus of all three, there's really nothing left to cook, and in fact, you never really learned how to cook to begin with. Nor advanced in it. You didn't acquire special tricks of a trade working within that trade. They're all in the engine already and you never needed them.

Similar things happened in music industry and movies. Entertainment turned from cultural and creative expression, or art, into fast food.

In a marketplace where the (perception is...) quality no longer makes the money, but the quantity, you want tools to produce quantity, not quality. Everything that happened in the real world of business... happens in the gaming business, but no one ever stops to think whether that's actually a good idea. A game is not to be treated as a product, but as a form of art. People appreciate movies not as products, but as works of art and creativity, just the same as music. And we have excellent detection mechanisms for things that have all the components to potentially be art, but really never got there at all. Art is not throwing things that work together. Art comes with a vision, an intent, a meaning, and manages to express that intent.

 

[Macro Device]

Start looking at the market.

You just explained me what I already know for like a couple ages. I wanted to know specifics of what I got no particular clue of.