Should full scene RT be abandoned? *spawn

[Nesh]

I ve been reading about Mega Geometry and how RT works and I get the impression that in certain scenarios RT is pointless and may either be producing worse results and/or have more performance hit than necessary.

I understand that RT reflections are necessary on huge surfaces, uneven surfaces and reflections within reflections. But occasionally we have selected use of RT in the environments with the common mirror and puddles. Many games have ugly noisy reflections and/or performance hits just to show these while everything else is matte or have high roughness value thus do not use or need RT.

What's the point? RT already uses a BVH structure to calculate reflections which is intensive. Many games produce the worst mirror reflections ever that take time to update and he reflections show lower detail because the BVH is lower just to have RT ticked. Only a portion of the environment is often displayed too. As in the case of AW2 on PS5 Pro, distance and detail of RT reflections on puddles are ugly as fuck. As such often reflections look disgustingly uglier than the fake methods used to replicate them during the PS2/XBOX/DC/GC era. See MGS2.

Silent Hill 2 Remake for example doesn't even have reflections on small mirrors during gameplay.

In such cases instead of having to use an already intensive processing BVH structure they could have just as well have partial replicated models for some mirrors and puddles and have them "mirrored". Since they are duplications of the same models and don't have to calculate tracing on an already a BVH representation of the environment, there is less strain on VRAM and processing. You just do it directly by having it directly on the "reflective" area.

Until GPUs are powerful enough to apply globally tracing of every surface (path tracing?), current implementations of RT is just ruining than improving in many many games

 

[DavidGraham]

I ve been reading about Mega Geometry and how RT works and I get the impression that in certain scenarios RT is pointless and may either be producing worse results and/or have more performance hit than necessary.

It's not about that. In certain very dense geometry situations, if you use ray tracing the performance drop will be large if ray tracing is used 1:1 compared to geometry. You can use lower dense geometry in the BVH and get slightly worse RT quality but you don't drop as much in performance.

However, the end result is far higher in visual quality compared to non RT solutions.

This is mostly about shadows though in relation to Nanite. Reflections and global illumination didn't suffer as much.

Mega Geometry aims to fix that and make the performance drop not as severe, so you can use the dense geometry in the BVH.

But occasionally we have selected use of RT in the environments with the common mirror and puddles

Non ray traced reflections are either screen space or planar, neither work on curved surfaces, both costs lots of performance if done with high resolution.

Screen space in particular are horrible, they don't support mirrors as they can't handle off screen objects, they suffer from disocclusion artifacts, and disappear partially/completely with camera movement.

 

[Nesh]

It's not about that. In certain very dense geometry situations, if you use ray tracing the performance drop will be large if ray tracing is used 1:1 compared to geometry. You can use lower dense geometry in the BVH and get slightly worse RT quality but you don't drop as much in performance.

However, the end result is far higher in visual quality compared to non RT solutions.

This is mostly about shadows though in relation to Nanite. Reflections and global illumination didn't suffer as much.

Mega Geometry aims to fix that and make the performance drop not as severe, so you can use the dense geometry in the BVH.


Non ray traced reflections are either screen space or planar, neither work on curved surfaces, both costs lots of performance if done with high resolution.

Screen space in particular are horrible, they don't support mirrors as they can't handle off screen objects, they suffer from disocclusion artifacts, and disappear partially/completely with camera movement.

I didn't suggest planar or screen space reflections. I made a specific case of flat reflective surfaces such us mirrors and puddles. For that it is less intensive and produce far better results to use mirrored geometry than rely on ugly results such as screen space reflections, or low geometry BVH or use performance intensive RT solutions

 

[trinibwoy]

I didn't suggest planar or screen space reflections. I made a specific case of flat reflective surfaces such us mirrors and puddles. For that it is less intensive and produce far better results to use mirrored geometry than rely on ugly results such as screen space reflections, or low geometry BVH or use performance intensive RT solutions

Mirrors and puddles are planar surfaces. It’s not cheaper to rasterize the scene multiple times if there are lots of those on screen at once. It’s just not scalable.

 

[Nesh]

Mirrors and puddles are planar surfaces. It’s not cheaper to rasterize the scene multiple times if there are lots of those on screen at once. It’s just not scalable.

It depends on the size of the reflection and area. You don't always need a duplication of the whole environment at the same detail to produce cheaper and better results.

 

[pcchen]

It depends on the size of the reflection and area. You don't always need a duplication of the whole environment at the same detail to produce cheaper and better results.

The thing is that you can't be sure before hand whether that's enough. Obviously if you put some limits on the scene you can do something like that, such as limiting the engine to just draw at most one mirror at a time. If in all your scenes you only have at most one mirror per room, that'd probably be ok, but for some it can be too limiting.
The same goes for outdoor water surfaces. It can be even more difficult to limit one pool per viewpoint. Generally it's better (and cheaper) to just use a technique which works in all situations, rather than having some arbitrary limitations.

 

[trinibwoy]

Generally it's better (and cheaper) to just use a technique which works in all situations, rather than having some arbitrary limitations.

People often say “well this and that game had cube maps or mirror reflections and looked great”. However those are not scalable solutions that can be applied in a general way to any content. That’s one of the benefits of RT. It is inherently scalable and generic. The only way to convince some folks is to go back in time and ask them to create a raster version of a game built around RT. We’re constantly going to get rose tinted nostalgia during the transition period.

 

[Cyan]

Generally it's better (and cheaper) to just use a technique which works in all situations, rather than having some arbitrary limitations

the performance should increase accordingly, and it's going to take a lot of time. The rumours about the RTX 60XX have started...

For the first time EVER, the rumour mill surrounding the next-gen GPU lineup isn’t kicking off with the usual “2× performance” claims. Frankly, given the current trajectory of silicon scaling and power efficiency bottlenecks, I have no clue what to expect in two years .

If AMD and Intel want to disrupt NVIDIA’s dominance, they can introduce a competitive architecture with a compelling perf-per-watt metric. This means not just raw rasterization gains but also efficiency-focused advancements that allow them to match NVIDIA’s performance while keeping power draw in check.

To summarize, the next major GPU iteration—expected in two years—will feature a modest 10% uplift in rasterization performance and a 20% boost in RT capabilities. VRAM capacity should increase (probably for Founders Edition SKUs), alongside an updated DLSS iteration and Frame Generation refinements. The transition to a 3nm fabrication node should theoretically improve power efficiency, but it may also be the limiting factor behind the relatively minor 10% rasterization gain due to clock-scaling constraints and diminishing returns in architectural optimizations.

NVIDIA's next-gen GeForce RTX 60 rumors begin: Rubin GPU, more VRAM, DLSS 5, TSMC 3nm node

PC gamers can't even buy NVIDIA's new GeForce RTX 5090 or RTX 5080, and now we've got the hype train for next-gen RTX 60 series GPUs. Let's go!

www.tweaktown.com

 

[Nesh]

People often say “well this and that game had cube maps or mirror reflections and looked great”. However those are not scalable solutions that can be applied in a general way to any content. That’s one of the benefits of RT. It is inherently scalable and generic. The only way to convince some folks is to go back in time and ask them to create a raster version of a game built around RT. We’re constantly going to get rose tinted nostalgia during the transition period.

It has nothing to do with nostalgia.
If you have an environment that is overall reflective sure RT is an easy way to uniformly add reflections.
But there are many environments that are using this costly method for the simplest effect when everything is have high roughness except a mirror. Or just a paddle. For example lets take mirrors. You dont even have to go that far back.
In many occasions they are producing worse results using this "advanced" RT effect.

Mirrors looked like mirrors before using such a simple technique.

Now RT for such a basic effect that was never a problem gives as the ugliest results

This was a generation ago without RT



This is Callisto Protocol not having the effect without RT on, and with RT we get an ugly result



This is Alan Wake 2 mirrors. A similar mirror that looked perfect on Beyond 2 Souls on PS3.



That result using a very costly on performance method. Not worth it for that case.

 

[pcchen]

I think this is again comparing a specific scenario to a more generic solution. I mean, they are not really mutually exclusive. You can still use RT with fake planar mirror if you really want to. It's just that the game developers don't do that for some reason.
The idea of using RT is, now since you already did BVH, you can do many related effects with relatively low extra cost. Yes, the initial cost is high, but it's now possible to achieve something that's very costly without it.
As mentioned before, if you have multiple mirrors or curved mirrors, it's very difficult and costly to do with traditional methods. The same goes to shadows from multiple light sources. Shadow maps are actually quite expensive.
Of course, we are currently in a "transitionary period" where games have to support hardwares with no RT or limited RT capabilities, so games can not be designed to take full advantage of RT. However, I expect that once RT capabilities become more common and more games require good RT capabilities this will change.

And yes we have to start somewhere. If people only do RT when all GPU have good RT capabilities, then we'll likely to see no RT at all, because why implement a feature in hardware when hardly no game uses it?

 

[Pjotr]

It mainly a effort thing. You can still use planar reflections if you want (even UE has this option) but it gives constrains on level layout (if you really going the duplicate geometry route that is). It might also have performance penalties, and even if it's not as severe as RT you will probably end up using RT in the game anyway so those performance penalties might stack. Here is a conversation about this where a remedy programmer who also worked on sifu (which uses planar reflections) shares some insight:

https://twitter.com/x/status/1767553164599869542

 

[Dictator]

This is Alan Wake 2 mirrors. A similar mirror that looked perfect on Beyond 2 Souls on PS3.

View attachment 13062

That result using a very costly on performance method. Not worth it for that case.

This image of this mirror in Alan Wake 2 is posted quite often, but the reason for that mirror looking distorted is not because "RT bad" it is because the mirror has material assigned to it which makes the light bend that way.
Alan Wake 2 has perfect mirrors (see below), but for some reason, the game art designers did not want that mirror to be a perfect one. It is not a technical problem or something.

For some reason the game art people wanted the reflective surfaces to have this kind of normal mapped detail on it:

 

[Shifty Geezer]

This image of this mirror in Alan Wake 2 is posted quite often, but the reason for that mirror looking distorted is not because "RT bad" it is because the mirror has material assigned to it which makes the light bend that way.
Alan Wake 2 has perfect mirrors (see below), but for some reason, the game art designers did not want that mirror to be a perfect one.

For some reason the game art people wanted the reflective surfaces to have this kind of normal mapped detail on it:

I'm guessing the distortion is plot related or in some way thematic. Indeed, counter to Nesh's point, this is better achieved with RT than planar reflections. Planer could add a distortion shader on top but it wouldn't be correct reflection.

 

[OlegSH]

I'm guessing the distortion is plot related or in some way thematic. Indeed, counter to Nesh's point, this is better achieved with RT than planar reflections. Planer could add a distortion shader on top but it wouldn't be correct reflection.

It's simpler than that. Typically designers do this to pure mirrors to hide the very obvious cubemap and SSR artifacts. Without the bumpy or scratched mirror surfaces, you would notice the ugly, fat cubemap texels and SSR artifacts in no time.

 

[Nesh]

Ok

This image of this mirror in Alan Wake 2 is posted quite often, but the reason for that mirror looking distorted is not because "RT bad" it is because the mirror has material assigned to it which makes the light bend that way.
Alan Wake 2 has perfect mirrors (see below), but for some reason, the game art designers did not want that mirror to be a perfect one. It is not a technical problem or something.

For some reason the game art people wanted the reflective surfaces to have this kind of normal mapped detail on it:

I get it if it has a purpose for AW2. But in order to run such HQ reflections you usually need expensive hardware.
Without that, even simple mirrors, as in the case of Callisto Protocol produce either ugly results or without RT on you dont even get a mirror.

What I feel about RT is that it got shoehorned in many games without the technology being up to the task, which at the end for many cases, even for RT supported hardware, the final result sometimes it is not that much noticeable or sometimes produces worse results (either through low quality reflections or with high performance cost), except for those that got the latest and most expensive hardware. For consoles especially (and first gen RT supporting GPUs) this has been a peculiar case of either sacrificing too much in performance with mixed results or missing the effects altogether making difficult to choose which option is best to run the game

Maybe the best argument is we got to start from somewhere. Sure. But at least until the HW is powerful enough, in a given game it makes sense to use RT in some areas and in some other areas we just aren't there yet. It is why many people, haven't been convinced nor see much a difference in many games. Especially in the beginning of RT introduction

We are probably only getting somewhere now with the high end hardware and new software solutions

 

[arandomguy]

It has nothing to do with nostalgia.
If you have an environment that is overall reflective sure RT is an easy way to uniformly add reflections.
But there are many environments that are using this costly method for the simplest effect when everything is have high roughness except a mirror. Or just a paddle. For example lets take mirrors. You dont even have to go that far back.
In many occasions they are producing worse results using this "advanced" RT effect.

Mirrors looked like mirrors before using such a simple technique.

Now RT for such a basic effect that was never a problem gives as the ugliest results

This was a generation ago without RT

View attachment 13060

Does that game have any other mirrors?

 

[Boss]

It has nothing to do with nostalgia.
If you have an environment that is overall reflective sure RT is an easy way to uniformly add reflections.
But there are many environments that are using this costly method for the simplest effect when everything is have high roughness except a mirror. Or just a paddle. For example lets take mirrors. You dont even have to go that far back.
In many occasions they are producing worse results using this "advanced" RT effect.

Mirrors looked like mirrors before using such a simple technique.

Now RT for such a basic effect that was never a problem gives as the ugliest results

This was a generation ago without RT

View attachment 13060

This is Callisto Protocol not having the effect without RT on, and with RT we get an ugly result

View attachment 13061

This is Alan Wake 2 mirrors. A similar mirror that looked perfect on Beyond 2 Souls on PS3.

View attachment 13062

That result using a very costly on performance method. Not worth it for that case.

Full scene RT imo is extremely wasteful especially in the era we're in with consoles and GPU's. Frankly, if you care about performance and craftsmanship, it's really hard to justify full scene RT as there are several instances where other methods offer a similar visual experience at a fraction of the cost. Right now, the RT implementations that yield the biggest gains in terms of cost-benefit analysis are RT global illumination and full path tracing. Stuff like RT shadows, reflections usually offer mediocre gains. To this day, there's only one game franchise that has wowed me with it's use of RT reflections and that is Spiderman since it's set in a city environment.

The rest are mostly a waste of performance as there are other techniques that offer a broadly "similar" visual experience. At some point in the future, entry level GPUs will be power enough to be that wasteful with performance but as of right now, it's just not worth it.

 

[DavidGraham]

But at least until the HW is powerful enough, in a given game it makes sense to use RT in some areas and in some other areas we just aren't there yet. It is why many people, haven't been convinced nor see much a difference in many games. Especially in the beginning of RT introduction

Practically 95% of RT implementations are provided as optional, you don't have to use them. You can simply work with the provided raster options. Which in the majority of reflection cases will be static cubemaps (so static mirrors), screen space (so ugly artifact ridden mirrors or no mirrors at all), or in some rare cases, planar (costs too much performance and doesn't look as good as ray tracing).

In Spider-Man Remastered and Spider-Man Miles Morales, the games use a mixture of static cubemaps (on buildings) and screen space (on puddles and water) reflections as their non ray traced options, the result is far less convincing than the ray traced alternative.

Spider-Man 2 did away with the raster option entirely, as it was time consuming for the developer, it only provides basic low res cubemaps for the raster path.

Some games smartly mix techniques to achieve higher performance, such as mixing ray traced and screen space reflections together, the base reflection will be screen space, unless there is an occluding or off screen object or a camera movement, in which case the ray tracing will kick in ... other techniques make the ray traced reflection the base one (so it reflects the geometry + basic shading) while providing the rest of the shading (lights, shadows .. etc) via screen space. You don't see these games many often because they require more time and effort from the developer.

 

[Remij]

How is graphics technology ever supposed to advance if you're stuck waiting for the hardware to be up to the task before trying it? You could very well easily argue that RT is as good as it is today because the hardware isn't up to the task. You hack solutions together until eventually the hardware is capable of doing so without it... then you move on to the next hacky solution. And I think they've done such a good job of it in such a short time that people forget just how intensive ray /path tracing anything is.. The fact that we're getting not just old games rendered with Path Tracing, but brand new current gen games with AAA visuals.. It's absolutely insane.

Nothing is being forced on anyone. I think the speed at which things are progressing is just as it's supposed to be. 10-15 years from now will be insane.

 

[Shifty Geezer]

Spider-Man 2 did away with the raster option entirely, as it was time consuming for the developer.

Just to throw the cat among the pigeons, automated tools should be able to raytrace a scene to determine what would need to go into a baked reflection and be able to create a suitable proxy. It should be possible to create a toolchain that'll allow a 'mirror' to be added and have the time-consuming stuff populated in a jiffy.

And I think they've done such a good job of it in such a short time that people forget just how intensive ray /path tracing anything is..

I'll argue that that's in part because of how RT was heralded. It was supposed to be photo real from the moment RTX 20xx appeared, with all sorts of preview and concept and demos ushering in a brave new world. The realities of the complexities of RT mean these targets were never realistic.

Nothing is being forced on anyone.

Not directly, but Nesh's point is that alternatives are being dropped. So where you might have a game that could use planar reflections to good effect, nowadays that option isn't present and you either have intensive RT requiring a high-end GPU or no reflection. If users could choose between which techniques they would prefer, there'd be no issues, but that's too much work to ask of the devs to create two (or more) different rendering paths and let the players decide.