• Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.

[Digital Foundry] Exclusive: Metro Exodus Enhanced Edition Analysis - The First Triple-A Ray Tracing Game

Andodalf

Banned
Ok? And what does that have to do with gameplay?

It makes things more immersive and cooler to look at.. but the same game could be made w/o it.
If I shine a light in the room, I shouldn't just see what it shines directly at. The room should be illuminated in areas that are exposed to bounces. It completely changes how the "hide and seek" works. You can actually have rooms be pitch black but still have bounce lighting.
 

IntentionalPun

Ask me about my wife's perfect butthole
If I shine a light in the room, I shouldn't just see what it shines directly at. The room should be illuminated in areas that are exposed to bounces. It completely changes how the "hide and seek" works. You can actually have rooms be pitch black but still have bounce lighting.
But GI can be done without RT.. or is it not possible for a moving light? I thought that was possible with light probes in engines like Unity?
 
Last edited:

VFXVeteran

Banned
But GI can be done without RT.. or is it not possible for a moving light?
Not what he's describing. GI without RT is like this:

Place a bunch of light probes inside the room and give them some arbitrary value that mimics the average color palette of the room itself. Turn off all of them save the ones that are near the flashlight.

That technique is both inaccurate and will never give the precise results of it's RT counterpart. If I go to a mirror in this room and shine the flashlight on my own armor for example, the light probes won't have that information stored so it won't be correct. Also the probes don't compensate for the diameter and area of the flashlight lens. The probes only give an approximate value, and you'd still have to run a screenspace ambient occlusion pass for checking for occlusion (that's why objects look flat shaded when not in direct sunlight). It suddenly gets way more difficult to art direct a scene this way.
 
Last edited:

Andodalf

Banned
Not what he's describing. GI without RT is like this:

Place a bunch of light probes inside the room and give them some arbitrary value that mimics the average color palette of the room itself. Turn off all of them save the ones that are near the flashlight.

That technique is both inaccurate and will never give the precise results of it's RT counterpart. If I go to a mirror in this room and shine the flashlight on my own armor for example, the light probes won't have that information stored so it won't be correct. Also the probes don't compensate for the diameter and area of the flashlight lens. The probes only give an approximate value, and you'd still have to run a screenspace ambient occlusion pass for checking with occlusion. It suddenly gets way more difficult to art direct a scene this way.

Additionally, even that level of Dynamic GI that is far more typically used for the Sun/Moon instead of something like a flashlight in a scene.
 

sinnergy

Member
It is because they changed the lighting tone so the comparisons are useless.
The graphics seems pretty similar with the same issues as 2019.

A fair comparison should be the scene with that new RT with the same lighting tone as in 2019 version... so we could really compare the implementation.
In a lot of scenes in the video the dark tone of 2019 fells better imo... the atmosphere changed.

But devs choose a way lighter tone maybe because it causes a big impression change.
Or that is the way lighting behaves .. just a thought …
What you can do , at least what I do , is set up your camera different in offline. ISO, Aparature etc … or add a tone mapper .

Or even real time , as I edit , my scenes nowadays in real-time RTX.
 
Last edited:
At first I was surprised by how well this runs, but then I thought, the GPU isn't having to process all those rasterized lights, so is not doing double the work like before. That should be why it works so well also on RDNA2 GPUs, shows how good they can be, the developer just needs to be mindful to not cross the limit of what it can calculate at a high framerate.
 

IntentionalPun

Ask me about my wife's perfect butthole
I appreciate the explanation VFXVeteran VFXVeteran .. but it's still just a more accurate version of something possible. Flashlight gameplay isn't something new.. you could make it better.. but it's still just.. a type of gameplay, that likely isn't used often.. not because of tech limitations, but because.. who cares? lol

It's mostly going to be used for better / more immersive graphics.. just like super fast I/O is going to mostly be used for better loading times or endless streaming worlds.

Which is fine; new tech is mostly bringing in gimmicks from a gameplay perspective, not true gameplay advancements. I'm still playing games that don't play much differently than the games I played a decade ago. They are just better/faster/prettier versions.. and I'm totally OK with that.
 
Last edited:

Zathalus

Member
Well that explains how the RT will be done on consoles. Using a native 4k image, and "Normal" RT will allow for the RT to only be calculated at 1080p, massively reducing the strain on the GPU. I'd imagine some of the other settings would be tuned down a bit to hit that target as well. I think this is proof of concept that the consoles can handle a full RT lighting model. PS6/Xbox Next would be the generation where the entire suit of RT features would be standard.

RNDA 2 is also not that far behind Ampere when properly optimised for either, which bodes well for the consoles going forward. That being said, RT will be limited to a single feature for most console games (Reflection, GI, Shadows, etc...).
 

Blond

Banned
This is the first time I can honestly say I see a clear, tangible difference in how RT changes the entire scene vs just "Oh look at this shadow in cyberpunk on a bench!" which just made the scene look different, not better. Spider-Man, Godfall, COD, etc I noticed no difference other than a jank framerate.
 

YCoCg

Member
but it's still just a more accurate version of something possible.
Yes, it's more accurate but from a game Dev pov it takes a fraction of the time! Look at the example in the video, toggling RT GI took pretty much a 1 second click and a 2 second transfer on their tools, compared to manually adding and adjusting hundreds of point lights to fake the effect, which in the example took around 28 MINUTES. That's a massive time saver and the results are more accurate.
 

ToTTenTranz

Banned
I watched the video. The full ~45min of it.

How guilty should I feel that I don't really find the RT-GI difference to be that big of a deal, nor do I find Metro Exodus an especially good looking game?
Same thing happened with Cyberpunk, by the way.

I can't help but feel like these are still just last-gen games with last-gen geometry, last-gen texture resolution and last-gen animations with prettier lights /lighting. I mean Minecraft RTX and Quake RTX are nice experiments but they're definitely not something I'd really want to play for their visuals.

Does the emperor have any clothes?
 
Last edited:

IntentionalPun

Ask me about my wife's perfect butthole
Yes, it's more accurate but from a game Dev pov it takes a fraction of the time! Look at the example in the video, toggling RT GI took pretty much a 1 second click and a 2 second transfer on their tools, compared to manually adding and adjusting hundreds of point lights to fake the effect, which in the example took around 28 MINUTES. That's a massive time saver and the results are more accurate.
I'm aware; this convo is about whether the RT actually creates "next gen gameplay." Or if the RT just enables other things; lack of need for baking lighting, easier to develop, better graphics, etc.
 

sinnergy

Member
I'm aware; this convo is about whether the RT actually creates "next gen gameplay." Or if the RT just enables other things; lack of need for baking lighting, easier to develop, better graphics, etc.
If someone would make a game about light and shadows … it would be about gameplay
 

IntentionalPun

Ask me about my wife's perfect butthole
If someone would make a game about light and shadows … it would be about gameplay

We are going in circles a bit here; not trying to be snarky.. but I already went down this path in this convo.

It's fine to disagree; and we all benefit if new techs really push "next-gen gameplay"... but I'm perfectly happy w/ better immersion, prettier graphics, etc. (and no load times with insane I/O, one of the other advancements of next-gen.) I don't think these are the techs that ae really going to create gameplay that wasn't possible before, they will enhance experiences for sure.

I wish people cared more about things like world simulation, not just from a graphical standpoint, because that's where I think games could actually truly innovate gameplay that wasn't possible before.

(while also continuing to advance graphics, leverage fast I/O, and the other next-gen advantages, that enhance games, but to me are largely independent of what is actually possible gameplay wise)
 
Last edited:

Shmunter

Member
It is because they changed the lighting tone so the comparisons are useless.
The graphics seems pretty similar with the same issues as 2019.

A fair comparison should be the scene with that new RT with the same lighting tone as in 2019 version... so we could really compare the implementation.
In a lot of scenes in the video the dark tone of 2019 fells better imo... the atmosphere changed.

But devs choose a way lighter tone maybe because it causes a big impression change.
It’s obvious from the video you do lose a lot of control in how the scene is lit with RT. Once you have that crack in the roof of the cave and light floods in, you’re at the mercy of the simulation. In movies, it’s all about point lighting and post editing to achieve a dramatic scene, I imagine a combination of both needs to still be employed in games if artistic intent is to be maintained and not just hands off accuracy.
 
Last edited:

Verchod

Member
This is an interesting thread, that's getting bogged down/confused by the idea that RT is used to model how light works, or to make scenes look realistic. That's not the same thing. Real lights can be used to make scenes that wouldn't be natural.
 

assurdum

Banned
But GI can be done without RT.. or is it not possible for a moving light? I thought that was possible with light probes in engines like Unity?
I think many if not most of the recent past generation games used "prebaked" GI. If you take any previous ps360 game you notice it immediately how incredibly "darker" were in comparison.
 
Last edited:

IntentionalPun

Ask me about my wife's perfect butthole
It’s obvious from the video you do lose a lot of control in how the scene is lit with RT. Once you have that crack in the roof of the cave and light floods in, you’re at the mercy of the simulation. In movies, it’s all about point lighting and post editing to achieve a dramatic scene, I imagine a combination of both needs to still be employed in games if artistic intent is to be maintained and not just hands off accuracy.

I'm so confused by this comment. Lose control?
 
Last edited:

tmlDan

Member
Am I the only one who thinks it doesn't look that good? :S

It def looks much better than the OG version but overall Metro just looks okay, I personally think Baldurs Gate looks better and its still early access.

Lighting definitely looks more realistic but is it worth it for how taxing it is?
 
Last edited:

Rikkori

Member
It's incredibly advanced TAA and why people downplay it is beyond me.
Because it's not, it's the next step, yes, but the incredibly advanced step hasn't arrived yet. They tried to - with DLSS 1.0, but it ended up as a still birth.
DLSS is not JUST TAA.

Someone correct me if I am wrong.

No one said "just", but it's TAA nonetheless. Nor would I ever say "just" TAA because even every TAA is not the same. I tried to make it easy with the soda analogy but apparently even eli5 is a few years too advanced. 🤷‍♂️
In some alternate reality where Radeon 6000 series + PS5 had DLSS and NVidia didn’t, 100% of the people downplaying it now would instead be praising and hyping it to the heavens. I think that tells you all you need to know.
Some of us just know more than others. ;)
It’s not just TAA. They use the TAA’s motion vectors with their solution, that’s pretty much it.. both use motion vectors, that’s almost the only thing in common.

It’s like comparing an heuristic optimization algorithm like genetic algorithms vs a deep learning approach and then say “lol, both maths both same”.

Nvidia still use their neural network to train it on ideal images vs low resolution image on their super computer, and then implemented in a driver basis. It bitchslaps TAA’s heuristic approach and fixed almost all it’s inherent problems. Even this thread’s video has a very good comparison and DF does not mince words here, DLSS 2 is king.
No, they don't train it on ideal (i.e. ground truth 16K-64K) images, that was DLSS 1.0. Again, more misinformation from people so hyped to praise it but such low interest to understand what it really is.

Source (also has the NVD presentation):


That was light work, keep trying fellas.

smirk dust off GIF by Efteling
 
Last edited:

Shmunter

Member
This is an interesting thread, that's getting bogged down/confused by the idea that RT is used to model how light works, or to make scenes look realistic. That's not the same thing. Real lights can be used to make scenes that wouldn't be natural.
Problem is, where do you hide those light sources in a user controlled camera? In a movie, you frame whatever you like.

e.g.

images


images
 

Lethal01

Member
It’s obvious from the video you do lose a lot of control in how the scene is lit with RT. Once you have that crack in the roof of the cave and light floods in, you’re at the mercy of the simulation. In movies, it’s all about point lighting and post editing to achieve a dramatic scene, I imagine a combination of both needs to still be employed in games if artistic intent is to be maintained and not just hands off accuracy.

You can set up RT lights to work the same as the usual fake lights if you want though, you can make them no bounce light if that's what you are aiming for. there are some random effects that are easier to achieve by simply using the old method but the general idea of using fake lights can be achieved just as easily with RT. Also you can hide lights by making them literally invisible.
 
Last edited:

ethomaz

Banned
It is one of these examples the new RT version feels off... there is no way, no matter how much sun you have, you can’t see what is outside... plus the inside won’t be that bright.

The previous RT looks better IMO.
They need to tone down the new RT lighting... they forced too much to make it different.
 
Last edited:

IntentionalPun

Ask me about my wife's perfect butthole
Because it's not, it's the next step, yes, but the incredibly advanced step hasn't arrived yet. They tried to - with DLSS 1.0, but it ended up as a still birth.
I’m well aware of the differences. It’s still incredibly advanced TAA; and it’s a better implementation because it is far more realistic to train on a common model to recognize various parts of an image that would normally cause issues than train on the actual game. And that’s not really any less advanced just more efficient from an implementation standpoint.

And it works better... and is more performant.
 
Last edited:

Tqaulity

Member
What games do you think are really pushing impossible ideas though?

I think Ratchet is the example used.. but we always could have had instantly changing worlds if that was actually some gameplay element that people wanted. You'd just be doing it with not as good graphics as you could w/ high i/o.

I think the PS5 is a fantastic machine, but also think so far we are just seeing really smooth awesome graphics and insanely low load times. And I absolutely LOVE the low load times.. but I find games that are doing cool simulations that are years old to be more advanced from a gameplay perspective than what I've seen "super fast I/O" being presented as bringing.

Not meant to be combative.. honestly wondering what is a "next gen game" and really the why you think it's "next-gen" worthy?
For me, A "next-gen" game is one that leverages latest technology across all avenues to offer new or improved gaming experiences than what came before. By "all avenues" I mean CPU (i.e. simulation), graphics, sound, input, I/O, UX and everything else that can affect the gaming experience. I explain it in my largely ignored thread here: What Constitutes Next Gen

My point with this Metro update is that (while pretty) it doesn't do anything to change/advance the actual gaming experience. Without having anything else to compare it to, it wouldn't even be obvious what the differences are since other games have achieved similar effect using prebaked solutions.

I watched the video. The full ~45min of it.

How guilty should I feel that I don't really find the RT-GI difference to be that big of a deal, nor do I find Metro Exodus an especially good looking game?
Same thing happened with Cyberpunk, by the way.

I can't help but feel like these are still just last-gen games with last-gen geometry, last-gen texture resolution and last-gen animations with prettier lights /lighting. I mean Minecraft RTX and Quake RTX are nice experiments but they're definitely not something I'd really want to play for their visuals.

Does the emperor have any clothes?
You shouldn't feel guilty at all. It's important to keep perspective and remember that evaluating graphics is completely subjective. Just because the numbers, algorithms, and technology is advanced doesn't mean that the end results will leave a positive impression on everyone. Your point about the game still be a last gen game at heart echoes my sentiments exactly.
 
Last edited:

IntentionalPun

Ask me about my wife's perfect butthole
For me, A "next-gen" game is one that leverages latest technology across all avenues to offer new or improved gaming experiences than what came before. By "all avenues" I mean CPU (i.e. simulation), graphics, sound, input, I/O, UX and everything else that can affect the gaming experience. I explain it in my largely ignored thread here: What Constitutes Next Gen

My point with this Metro update is that (while pretty) it doesn't do anything to change/advance the actual gaming experience. Without having anything else to compare it to, it wouldn't even be obvious what the differences are since other games have achieved similar effect using prebaked solutions.
Thanks for the explanation, I'll take a look.
 

DeepEnigma

Gold Member
It is one of these examples the new RT version feels off... there is no way, no matter how much sun you have, you can’t see what is outside... plus the inside won’t be that bright.

The previous RT looks better IMO.
They need to tone down the new RT lighting... they forced too much to make it different.
I agree with this, which is where they need to adjust it so it’s what your eyes would actually see being there, not what a camera’s imaging sensors will see. Like they do in film, etc..
 

Lethal01

Member
I agree with this, which is where they need to adjust it so it’s what your eyes would actually see being there, not what a camera’s imaging sensors will see. Like they do in film, etc..

I think both are totally fine approaches, and the good news is they can mix and match. Blinding light coming from outside is a great tool even if it's unrealistic.
 

Lethal01

Member
We won't see another developer have the balls to do this for a while. Since 2018 4A games has been the chosen one for raytracing.

You may see it on PS5 exclusives. Ratcher and Clank has raytracing even in it's 60fps mode so technically that game also only works on Raytracing capable GPUs
 
Would this type of global illumination be possible on next gen consoles?

it’s really actually quite striking it’s overall effect on the graphics - gives it that prerendered look - and this is coming from someone who has not be too impressed with RT as of yet
 
Last edited:

Mister Wolf

Gold Member
You may see it on PS5 exclusives. Ratcher and Clank has raytracing even in it's 60fps mode so technically that game also only works on Raytracing capable GPUs

I doubt we see another game whose whole lighting engine is raytraced. They wont want to dedicate the resources to it. 4A games is obsessed with raytracing more than anyone else. Especially not a Sony Studio game. I would rather R&C had raytraced emissive textures but there obviously is a reason why it doesn't. Imagine if all the light producing stuff in that game actually lit the environment in that 15 min demo. I wish Halo Infinite would do it for PC but I know they wont.
 
Last edited:

VFXVeteran

Banned
It’s obvious from the video you do lose a lot of control in how the scene is lit with RT. Once you have that crack in the roof of the cave and light floods in, you’re at the mercy of the simulation. In movies, it’s all about point lighting and post editing to achieve a dramatic scene, I imagine a combination of both needs to still be employed in games if artistic intent is to be maintained and not just hands off accuracy.
That's why you have render passes. You render out things like texture, AO, specular, diffuse, etc.. and then do composite operations on those to tweak your scene. It's much better and faster to do it that way than to try placing fill lights, managing SSAO (which is completely inaccurate), and using shadow maps to try and get good results + still having to deal with render passes. It's truly a huge time sync.
 

Rikkori

Member
I’m well aware of the differences. It’s still incredibly advanced TAA; and it’s a better implementation because it is far more realistic to train on a common model to recognize various parts of an image that would normally cause issues than train on the actual game. And that’s not really any less advanced just more efficient from an implementation standpoint.

And it works better... and is more performant.
Gotta dream bigger than that, bro. Here's what I'd consider that advanced step and not just incremental:





 
Last edited:

hlm666

Member
Would this type of global illumination be possible on next gen consoles?

it’s really actually quite striking it’s overall effect on the graphics - gives it that prerendered look - and this is coming from someone who has not be too impressed with RT as of yet

Its coming for ps5/xsx so yes, they might have to use the 1/2 or 1/4 res RTGI setting but it's happening.
 

YCoCg

Member
Great! Guess faking it will still have a place to achieve desired lighting, and rightfully so.
Well yeah, of course, just as you pointed out, movies and TV shows don't exactly just rely on daylight.

Ray Traced GI here is being put to use for a realistic style but there's no reason why it can't be combined with stylistic stuff either, if anything the old motto comes into place "it's easier to work backwards on some things". If anything working with RT GI and Lighting will make game Devs think even more like movie/TV show makers as they'd be looking to include box lighting, filters, masking, etc.

Just that would take 15 minutes to work out compared to say 45-50 minutes for example work wise.
 

Lethal01

Member
I doubt we see another game whose whole lighting engine is raytraced.
They wont want to dedicate the resources to it. 4A games is obsessed with raytracing more than anyone else. Especially not a Sony Studio game. I would rather R&C had raytraced emissive textures but there obviously is a reason why it doesn't. Imagine if all the light producing stuff in that game actually lit the environment in that 15 min demo. I wish Halo Infinite would do it for PC but I know they wont.

I'm not talking about how good the ray tracing is though.

What they "had the balls" to do, is making a version of a game built around the raytracing without bothering to set it up to work without said raytracing. It is indeed likely that you see games on PS5 and maybe XBSX that do the same thing even if they are using less RT effects, Since as soon as you decide to make a game exclusively for those systems you already know that you are limiting to a player-base that are all using RT capable hardware.

Funny enough Ratchet and Clank is an example of a game that is already doing it since you don't have the option to turn off the Rt in that game.

But I agree with being doubtful that we get someone else who does this for a AAA "PC" game though. I would be extremely happy to be wrong.
 
Last edited:

Mister Wolf

Gold Member
I'm not talking about how good the ray tracing is though.

What they "had the balls" to do is make a version of the game built around the raytracing and not even bother to set it up to work without that raytracing. It is indeed likely that you see games on PS5 and maybe XBSX that do the same thing even if they are using less RT effects, Since as soon as you decide to make a game exclusively for those systems you already know that you are limiting to a player base that are all using RT capable hardware.

Funny enough Ratchet and Clank is an example of a game that is already doing it since you don't have the option to turn off the Rt in that game.

But I agree with being doubtful that we get someone else who does this for a AAA "PC" game though. I would be extremely happy to be wrong.

Well it seems we were discussing two different things. The marvel of this game is specifically its entire lighting being raytraced. You wont see it again any time soon.
 

Lethal01

Member
Well it seems we were discussing two different things. The marvel of this game is specifically its entire lighting being raytraced. You wont see it again any time soon.

Thought that may be the case but it could have gone either way.

I'd bet on you being right if I were forced to. But I can also see another game giving the option for raytraced GI, Reflections, Emissions, and Shadows especially if you include indie games, the difference is that they will probably still have the option to use the old lighting methods so they won't be totally built around it and won't look as good artistically.
 

Neo_game

Member
Well that explains how the RT will be done on consoles. Using a native 4k image, and "Normal" RT will allow for the RT to only be calculated at 1080p, massively reducing the strain on the GPU. I'd imagine some of the other settings would be tuned down a bit to hit that target as well. I think this is proof of concept that the consoles can handle a full RT lighting model. PS6/Xbox Next would be the generation where the entire suit of RT features would be standard.

RNDA 2 is also not that far behind Ampere when properly optimised for either, which bodes well for the consoles going forward. That being said, RT will be limited to a single feature for most console games (Reflection, GI, Shadows, etc...).

We already knew this at launch. I think it was GT7 trailer in which the RT was running low res. Also WDL RT settings from the ini file. 1080P for PS5 SX and 720P on the SS. 4A devs said RTX 2070 and f 6800XT will be able to do 1080P 60fps? That seems pretty bad for 6800XT. So I do not think much should be expected from the consoles.
 
Its coming for ps5/xsx so yes, they might have to use the 1/2 or 1/4 res RTGI setting but it's happening.


They wont "might have to" use the noisy quarter res. They will definitely use it and much more. Considering the 6800XT, more than twice as fast than whats in a PS5 goes bellow 60, at 1440p, with a top of the line cpu. People should not expect on ps5 the same game they watched in this video, visually. Considering this is a pc only release for now, it seems they need to tweak quite a bit over there
 
Top Bottom