Digital Foundry Article Technical Discussion [2023]

[Dampf]

Holy shit. It's so much more atmospheric. Absolutely worth the performance hit.

Going to enjoy this in 2 years.

 

[Nesh]

for sure, I think artistic design and lighting needs a pass now that it's switched over to RT.
I don't necessarily agree that things being too dark means it looks bad however. The biggest difference between RT and Not RT will come down to one being static and one being dynamic.
Yes in screenshots, or stills, nothing is changing, nothing is dynamic, so we can appreciate artist vision on lighting. I think when thing become dynamic, then we can appreciate artist detail, and we can appreciate lighting direction.

For the areas that are too dark, yes, it's hard to see, but perjhaps when you get close, they light a cigarette lighting up the surrounding pillars they are around. Stuff like that, that makes it good.
Having that contrast is something different, and as these new technologies come into play, artist will be able to play with light in more ways than just setting the mood. Now they can literally have us have gun fights in complete darkness where the gun flares are giving us vision of where things are. Blowing holes through walls of a dark room for light to peek in and light the room dynamically.

When you can't change dark to bright and bright to dark, not setting these conditions could seem bad. But if you can let players or dynamically alter these conditions while players are playing, it goes from bad, to awesome. And I think that aspect is lost, if you don't consider the dynamic nature of light as an interactive tool in our interactive medium.

So while I fully agree with your point of view, I'm still going to agree with DF here and say RT = Good!

Every application I ve seen of RT in actual games was underwhelming especially for the performance cost. But this....this is on a completely different level. The results are transformative.
This is what we really needed for next gen. Unfortunately this rquires a hefty PC and consoles cant even dream approaching this kind of quality

 

[iroboto]

Every application I ve seen of RT in actual games was underwhelming especially for the performance cost. But this....this is on a completely different level. The results are transformative.
This is what we really needed for next gen. Unfortunately this rquires a hefty PC and consoles cant even dream approaching this kind of quality

I honestly think we'll get there if Metro EE is any indication. Not Cyberpunk levels, but at least, to the point that we no longer need to do baked lighting. I think the consoles are capable enough to handle that much, and I think that would be a sufficient first step.

 

[Nesh]

I honestly think we'll get there if Metro EE is any indication. Not Cyberpunk levels, but at least, to the point that we no longer need to do baked lighting. I think the consoles are capable enough to handle that much, and I think that would be a sufficient first step.

Fingers crossed.

 

[Silent_Buddha]

I remember the ME:EE discussion. FWIW I completely disagreed with your opinion there. To me the RT version looked better aesthetically. (And of course it was strictly better from a physical correctness point of view as well.) But that's a subjective thing, and your opinion is just as valid as mine.

I think we all agree that changing the lighting model *ideally* needs an artist do-over. And from the Cyberpunk video it does appear that they did some "light-touch" nips and tucks. But a complete do-over is probably too expensive. Still, from the videos at least, it looks like this specific game looks GREAT most of the time even without such a do-over. I am sure there will be exceptions where it looks worse. Overall, based on the video and @Dictator's commentary it seems to be an overwhelming win. So I don't understand all the outrage at the enthusiasm being expressed.

Whatever the lighting model is, artists will always need to add nips and tucks to get the lighting to match a creative vision. The question is, what's a better baseline to start from? A broken-ass system of hackery that makes everything glow-in-the-dark, or a physically correct RT baseline? And that's my problem with arguments that say "RT lighting needs artist work". Yes, it's true. But it's implicitly drawing a false equivalence (perhaps unintentionally) between a broken rasterized baseline and a ridiculously superior RT baseline.

Yes, what looks good is often subjective and lighting is no exception. I also respect the opinions of people that are perfectly fine with how the current state of RT looks and even if they might prefer the realistic versus film, TV, print, advertising, studio photography, comic, etc. look.

But it doesn't hurt to have other opinions presented as well. It's always nice to come across posts like these that are rationally presented versus some of the sometimes rabid (at least that's how they sometimes come across) proselytizing of a certain viewpoint that we often come across.

I mean, I've been 100% glad that people have been perfectly happy with what their RT hardware provides and the current mini-steps that some developers are making WRT RT even if it still isn't at a point where I'd find it "good enough" for the performance hit incurred on anything but the most powerful hardware without (again, IMO) the use of reconstruction that isn't stable enough in motion most of the time.

There's room for dissenting opinions without people getting snarky as often seems to be the case.

Regards,
SB

 

[trinibwoy]

Every application I ve seen of RT in actual games was underwhelming especially for the performance cost. But this....this is on a completely different level. The results are transformative.
This is what we really needed for next gen. Unfortunately this rquires a hefty PC and consoles cant even dream approaching this kind of quality

I think that comes down to the complexity and variety of the lighting. CP has a lot of superficial light sources that interplay with complex geometry. It's a best case scenario for showcasing pixel perfect GI and shadows. Psycho RT with just one bounce of RT sun GI already looked really good. In games where the sun is the primary light source it's a lot easier to fake a good result (baked or otherwise).

 

[Dampf]

Something I'm very curious about is the performance impact from hybrid Raytracing here:


@Dictator Am I looking at this the right way? It appears the performance impact from enabling Psycho Raytracing is ridiculously small compared to previous architectures like Ampere and Turing. On these architecture's you would be looking at half the performance or worse. But on Ada it seemingly appears to impact performance by just around 20%.

Could this be because the patch enables SER for the Lovelace architecture which was not enabled previously and is not hardware accelerated on older architectures? If that's the case, then SER is pretty much the real deal and could come in handy in future hybrid raytracing games.

 

[ethomaz]

From somebody that got really disapointed with the RT lighting in Metro (that looks worst than the previous backed in most cases making the atmosphere too bright for a game supposed to be dark) I can understand the @Rurouni opinion.

Having RT doesn't mate it looks better by default... RT can look worst than rasterization in a lot of cases.

I need to try Cybepunk first to have a better opinion.

 

[Jupiter]

Something I'm very curious about is the performance impact from hybrid Raytracing here:

View attachment 8718
@Dictator Am I looking at this the right way? It appears the performance impact from enabling Psycho Raytracing is ridiculously small compared to previous architectures like Ampere and Turing. On these architecture's you would be looking at half the performance or worse. But on Ada it seemingly appears to impact performance by just around 20%.

Could this be because the patch enables SER for the Lovelace architecture which was not enabled previously and is not hardware accelerated on older architectures? If that's the case, then SER is pretty much the real deal and could come in handy in future hybrid raytracing games.

Maybe but maxed out raster settings also had a bad performance/visual ratio.

Raster becomes more and more inefficient. You might as well switch to RT for reflections before using Ultra SSR settings.

 

[trinibwoy]

@Dictator did CDPR mention how many GI bounces they're doing for each RESTIR sample? It would be good to see how Psycho and Overdrive differ in this scene for example.

 

[Scott_Arm]

Have you played it? From watching videos I agree the art design isn’t great but maybe I would like it more when actually playing. If I only saw videos of saints row 3/4 I wouldn’t like it either but actually playing those games it was a different story.

I haven't played it. I've only really looked at videos exploring the graphics. Gameplay wise it didn't interest me. I think my main issue is the city looks very sparse. I was expecting a game that would look a lot darker and grimier. Urban decay mixed with high tech. A huge chasm between rich and poor. It kind of looks like that, but overall the world looks way too sterile for me, and way too empty in general. It looks like kind of clean and modern, with the occasional garbage bag and loose newspaper thrown around.

 

[Flappy Pannus]

Jesus, pretty stunning. The lighting of the NPC's in particular has always stood out so starkly to me in CP 2077, this completely fixes that.

 

[iroboto]

. A huge chasm between rich and poor. It kind of looks like that, but overall the world looks way too sterile for me, and way too empty in general.

no sarcasm on this one: this is why we need more VRAM and a better IO system. If you want to pile garbage everywhere, and just layer tons of crap, the texturing costs start to rise, you'll need a streaming system to account for all of these little objects and trash about.

Of course, if the papers are static and can't be moved or shot, I guess that's just world modelling static papers onto everything ie: The Division 1.

 

[Nesh]

I haven't played it. I've only really looked at videos exploring the graphics. Gameplay wise it didn't interest me. I think my main issue is the city looks very sparse. I was expecting a game that would look a lot darker and grimier. Urban decay mixed with high tech. A huge chasm between rich and poor. It kind of looks like that, but overall the world looks way too sterile for me, and way too empty in general. It looks like kind of clean and modern, with the occasional garbage bag and loose newspaper thrown around.

Yeah we need something closer to Blade Runner or Demolition Man or Robocop.
Most renditions of the future overdo it with the sterile clean look. Its one of my gribes with how the art evolved in the Wipeout series as well.
These movies and the original Wipeout games were more believable because the worlds communicated the element of time/decay, weather and a living breathing world.

 

[T2098]

Whatever the lighting model is, artists will always need to add nips and tucks to get the lighting to match a creative vision. The question is, what's a better baseline to start from? A broken-ass system of hackery that makes everything glow-in-the-dark, or a physically correct RT baseline? And that's my problem with arguments that say "RT lighting needs artist work". Yes, it's true. But it's implicitly drawing a false equivalence (perhaps unintentionally) between a broken rasterized baseline and a ridiculously superior RT baseline.

Agreed, and one thing that sometimes gets glossed over is that having your artists start the design process with the assumption that light will largely obey the law of physics and be *consistent* is itself a huge win. At least some parts of materials and art design start before the engine is finalized or even feature-complete, and having to re-do it all because at some point mid-project the engine changed and swapped out a bunch of rasterization 'approximations' is never fun.

We're now at the point where artists on next-gen projects can create everything with the broad assumption that light will behave in a consistent way, and making it look better on hardware 10 years from now is relatively straightforward and shouldn't break the original vision of the scene. Turn up the number of bounces evaluated, increase the number of rays per pixel, swap in a much better and high quality denoiser, etc. and the original vision will be retained.

 

[Dr. Nick]

Yes. While obviously I use RT/path tracing, there are plenty of fakes that I do to please the client. Although in general realistic is preferable, but within that realism, often times client ask to tweak something that if you understand physics, it is obvious that it doesn't behave like that in real life. The easiest example is light/dark part like I mentioned before. Other thing would be like having the light bright enough to illuminate a room but doesn't want to over exposure object that are too close to the light. A lot of tricks being used to get to the final image that can satisfy the client (either in post or in the model/lighting setup). Ideally of course I should let physics do its thing and say to the client that "It's physics!", but unfortunately something that is art doesn't really care about how accurate you're trying to simulate reality.

I don't entirely agree, but I do believe you have a point. Hero lighting sometimes has to be used because if the scene is lit correctly, you can hardly make out specific details. I think certain fighting games use hero lighting because very realistic lighting might have the character blend into the background too much, making it harder to see certain moves. I think Alex should be able to say how incredible something looks, even if the lighting might not be appropriate in all situations.

 

[dobwal]

for sure, I think artistic design and lighting needs a pass now that it's switched over to RT.
I don't necessarily agree that things being too dark means it looks bad however. The biggest difference between RT and Not RT will come down to one being static and one being dynamic.
Yes in screenshots, or stills, nothing is changing, nothing is dynamic, so we can appreciate artist vision on lighting. I think when thing become dynamic, then we can appreciate artist detail, and we can appreciate lighting direction.

For the areas that are too dark, yes, it's hard to see, but perjhaps when you get close, they light a cigarette lighting up the surrounding pillars they are around. Stuff like that, that makes it good.
Having that contrast is something different, and as these new technologies come into play, artist will be able to play with light in more ways than just setting the mood. Now they can literally have us have gun fights in complete darkness where the gun flares are giving us vision of where things are. Blowing holes through walls of a dark room for light to peek in and light the room dynamically.

When you can't change dark to bright and bright to dark, not setting these conditions could seem bad. But if you can let players or dynamically alter these conditions while players are playing, it goes from bad, to awesome. And I think that aspect is lost, if you don't consider the dynamic nature of light as an interactive tool in our interactive medium.

So while I fully agree with your point of view, I'm still going to agree with DF here and say RT = Good!

RT doesn't eliminate the ability of devs to stage lights in a scene. However, devs are still encumbered by non-RT hardware or RT hardware that isn't performant enough so they have to devote resources to multiple lighting solutions. Once performant RT hardware becomes more prevalent, I imagine more devs will sit down and really use RT in a way thats more than just being physically correct.

 

[Scott_Arm]

no sarcasm on this one: this is why we need more VRAM and a better IO system. If you want to pile garbage everywhere, and just layer tons of crap, the texturing costs start to rise, you'll need a streaming system to account for all of these little objects and trash about.

Of course, if the papers are static and can't be moved or shot, I guess that's just world modelling static papers onto everything ie: The Division 1.

I'd be curious to see how well UE5 would handle this.

 

[DavidGraham]

I am just glad that PC hardware can finally stretch it's legs with next gen visuals of it's own, independent of the reliance on console ports and the likes. It really remins me of the good old Crysis and Far Cry days. I miss those days, and I miss the enjoyment I get from interacting with state of the art graphics.

 

[Dampf]

I am just glad that PC hardware can finally stretch it's legs with next gen visuals of it's own, independent of the reliance on console ports and the likes. It really remins me of the good old Crysis and Far Cry days. I miss those days, and I miss the enjoyment I get from interacting with state of the art graphics.

The lighting is next gen now, but the rest isn't. Although it gets fairly close with HD Reworked textures. But still, asset quality, particles and draw distance is lacking compared to Nanite stuff and TLOU.

I wouldn't call that a next gen experience just yet.