Digital Foundry Article Technical Discussion [2023]

Status
Not open for further replies.
Can you show the screenshots of this inconsistency? The game looks damn near perfect in the jungle.
I posted a few earlier in the thread. There's sometimes a lack of even stuff like AO in the GI representation. It's especially stark in the interiors (which often even miss direct occlusion/shadowing), but it shows up in the exteriors as well.

That said, I'm all for Avatar getting put as #1, specifically for the achievement of how it looks at its performance level on consoles as Oliver called out. I just don't think it is really doing anything mind-blowing in the lighting department when it scales up on a high end PC. I believe folks when they say that there's raytraced GI here, but I still maintain that I would not be able to guess that based on the visuals alone.

It seems like overall a lot of this discussion is becoming more subjective over time, which is probably to be expected. As technology removes more of the obvious visual limitations people's impressions of artistic choices start to dominate the conversation more. Two people looking at the same two games can have wildly different impressions and preferences now. I think that's fine and obviously makes for good conversation as long as folks remain civil and aware of the base subjectivity. Arguably though it might mean stuff like numbered best of lists lose a bit of utility going forward, but I know they are important for the youtube algorithm and functionally if they just bring up a bag of different things to talk about that's fine too.
 
@Andrew Lauritzen I think people generally have a hard time separating art from technology (myself included). Does best graphics mean best graphics technology or best graphics presentation which would be art leveraged by technology? For Avatar I'd have different answers for each.
Indeed and it's ultimately subjective because people are effectively answering different questions, or at the very least with different weights on different things. I think that's fine, and it's part of what makes the discussion between different folks interesting as long as no one gets too serious about it :D
 
@Andrew Lauritzen I think people generally have a hard time separating art from technology (myself included). Does best graphics mean best graphics technology or best graphics presentation which would be art leveraged by technology? For Avatar I'd have different answers for each.
And then even if you count art style, how much does it factor into the final verdict? It's ultimately all subjective. To me, best graphics are the same as best visuals, ie: how good is it when I look at it? Animations, art direction, style, technical prowess, etc, all matter to me. If we were talking strictly most advanced games graphically, then that'd be a different story. Art direction/style doesn't factor very much in there.

I do think it's important to celebrate games that don't necessarily push technical boundaries but still have a lot of visual flair. games like Guilty Gear, Ori, Hi-Fi Rush, or Streets of Rage 4 shouldn't overlooked simply because they don't have ray tracing or use the latest and greatest rendering technologies.
 
This is less of an issue that DF should focus on. In my opinion a game like Guilty Gear has no place in a list of the most graphically impressive games. I wouldn't include the Switch games either.

I would rather include a game like ARK 2 in this list. That game also has similarities with Crysis.

Nevertheless, art is also important. I find Control particularly artistic.
 
Last edited:
There's also stuff like this which I think just looks bad. It looks wrong somehow. The rock faces look flat. Without the jungle in the distant background I could be convinced this was from a very old game.

View attachment 10338

It's hard (impossible) to make low geometry surfaces look good when combined with really good lighting. Basically, really good lighting, IMO, destroys/makes texture tricks that attempt to hide low geometry look really REALLY bad.

It's definitely a problem that Avatar has, although a lot people are rightfully wowed by it's lighting consistency and foliage density (A+), the low geometric complexity of the world, for me, is a huge letdown and just looks bad when combined with the generally good lighing.

Regards,
SB
 
It's hard (impossible) to make low geometry surfaces look good when combined with really good lighting. Basically, really good lighting, IMO, destroys/makes texture tricks that attempt to hide low geometry look really REALLY bad.

I'm sorry but this isn't at all true as clearly demonstrated by the likes of Lego Builders, Portal RTX, Minecraft RTX to name a few obvious ones.

It's definitely a problem that Avatar has, although a lot people are rightfully wowed by it's lighting consistency and foliage density (A+), the low geometric complexity of the world, for me, is a huge letdown and just looks bad when combined with the generally good lighing.

Again, the world is not "low geometry". It sports an insane amount of geometry. Likely one of the highest of any game released to date. Just because you can point to a random vine that's made of actual geometry with a few sharp angles in a jungle bursting with foliage which in any other game would have probably been a simple texture doesn't mean the game can be classified as low geometry.
 
Can you show the screenshots of this inconsistency? The game looks damn near perfect in the jungle.
 
You do know CP2077 Over Drive mode exists right? And that CP2077 has tones of low geometry surfaces.

Yah that’s the primary reason I don’t think CP2077 is as visually good as Alan wake 2 or Avatar (in some areas). Cyberpunk just looks kind of simple. It’s very clean with straight lines and not much clutter. The roads are very smooth and weirdly shiny. It looks like ps4 geometry and world detail with amazing lighting. Some npcs and areas look highly detailed and I think the expansion content looks better.
 
My order would be

1. Alan Wake
2. Avatar
3. Cyberpunk

Alan wake comes the closest to looking like its intended render target. The PT isn’t as good as Cyberpunk but the assets and general art quality are so much better it easily pulls ahead for me.
 
I'm sorry but this isn't at all true as clearly demonstrated by the likes of Lego Builders, Portal RTX, Minecraft RTX to name a few obvious ones.



Again, the world is not "low geometry". It sports an insane amount of geometry. Likely one of the highest of any game released to date. Just because you can point to a random vine that's made of actual geometry with a few sharp angles in a jungle bursting with foliage which in any other game would have probably been a simple texture doesn't mean the game can be classified as low geometry.

They explain it well the terrain in Avatar did not receive the same attention than the other part of the geometry because most of the time it is cover by vegetation.
 
Was watching the DF weekly where they were talking about FG and consoles and they bought up dynamic res. Would FG with DRS actually work? Would the resolution changing cause artifacts in the interpolation? If they can't I wonder which use cases a dev would pick drs over frame gen with a locked lower base res.
 
Avatar has scenes that just look weird to me. Pretty much a lot of the scenes with fog at night has raised blacks so all of the contrast gets lost. I'm actually not a person that likes super contrast (I played the Destiny 2 demo and it turned me off the game), but I think it's part of the reason why this game looks very flat. It's too low contrast.

View attachment 10337

There's also stuff like this which I think just looks bad. It looks wrong somehow. The rock faces look flat. Without the jungle in the distant background I could be convinced this was from a very old game.
This is the result of having to only sample a sphere of possibly low detailed environment sky (possibly spherical harmonics?). Since the objects are in shadow, you can't possibly shade them anything else other than environment sky. With no other bounced light being importance sampling the bright areas of the sky, it's probably no other solution.

Same thing here since it looks like the entire sky is just a constant haze of vanilla color. No direct lighting there and just using ray-tracing a constant color environment sky.

But then you get a ton of stuff that looks like this which looks fantastic. If you told me this screenshot was from a much newer and totally different game, I'd believe you.

View attachment 10339

Edit: I'll add that I haven't played the game, so I'm only going by what's provided. I've never seen it in HDR, so the presentation of this stuff
Direct lighting + indirect lighting showing proper contrast.

could look much different and better. There's a lot of variables. Overall Avatar is super technically impressive. I'd never doubt that. I just see certain scenes that I don't think look good, which can be as much about art direction as technology.
Indirect lighting isn't the best implementation without using path-tracing and importance sampling. It is what it is.
 
I posted a few earlier in the thread. There's sometimes a lack of even stuff like AO in the GI representation. It's especially stark in the interiors (which often even miss direct occlusion/shadowing), but it shows up in the exteriors as well.
Hmm..
That said, I'm all for Avatar getting put as #1, specifically for the achievement of how it looks at its performance level on consoles as Oliver called out. I just don't think it is really doing anything mind-blowing in the lighting department when it scales up on a high end PC. I believe folks when they say that there's raytraced GI here, but I still maintain that I would not be able to guess that based on the visuals alone.
Yea, having only single-cast RT GI isn't enough for good contrast varying quality illumination. That can only be achieved with importance sampling the environment. Importance sampling would bring out the brighter higher contrast sources into the indirect lighting part of the equation. I'm not sure if they only fired one ray against a fairly constant colored proxy geo or what. Of course importance sampling is a fundamental technique in path-tracing (which this game is not doing).
 
Yea, having only single-cast RT GI isn't enough for good contrast varying quality illumination. That can only be achieved with importance sampling the environment. Importance sampling would bring out the brighter higher contrast sources into the indirect lighting part of the equation.
It's quite likely that Avatar does use importance sampling. Importance sampling is a fundamental technique for decreasing noise for any GI implementation and is widely used across all real-time GI implemetions, both older ones like in Metro which people call RTGI and newer ones like in Cyberpunk Overdrive, which people call path tracing.

 
It's quite likely that Avatar does use importance sampling. Importance sampling is a fundamental technique for decreasing noise for any GI implementation and is widely used across all real-time GI implemetions, both older ones like in Metro which people call RTGI and newer ones like in Cyberpunk Overdrive, which people call path tracing.


From the DF interview:

“If I can explain it briefly in path tracing terms, what we do is a specific technique called "guided paths". Basically on the first ray that you hit, you evaluate the lighting analytically. So you don't just do complete Monte Carlo path tracing. But this is only there for the analytical lights. For the emissive surfaces, as Sasha was saying, we actually rely on the randomness of the rays. So that's why this can introduce more noise than analytics lights. But emissives do work and we fully support them.”

If I’m reading that correctly they’re importance sampling light sources but emissive geometry only contributes to lighting if a ray happens to hit the emissive surface on the first (and only) bounce.

It sounds extremely similar to Metro Exodus EE with an added screen space trace for further optimization.
 
Tracing rays explicitly towards emissive surfaces is usually referred to as next event estimation or direct lighting. I'm not familiar with any game which would handle emisse surfaces in a such manner as it's doubling the GI work and is expensive for real-time.

Importance sampling usually refers to exploiting some additional information to decide where to send more rays. The simplest form of it is BRDF importance sampling, where some directions get more rays based on the surface properties. If you imagine a mirror you don't want to trace rays in all directions uniformly. Instead you want to focus all your rays in the direction of the reflection.

BRDF importance sampling is how both Metro Exodus and Cyberpunk: Overdrive before the recent 2.1 update calculated GI.
Ultimately, though, we’ve ended up using ReSTIR DI mostly for the analytical local lights with uniform sampling, so the falloff issue wasn’t a problem anymore, and handled emissives and skylight with naïve BRDF sampling while tracing paths for the indirect light.

At the end of the day all techniques are similar. Some form of importance sampling, ray guiding, ray caching and a denoiser on top. Devil is in the details, which often can be only compared by reading papers and talks, as games have lots of tricks like simplified BVH, simplified materials for ray tracing or emulating GI with manually placed analytical lights, which hide strengths and weaknesses of GI methods.
 
It sounds extremely similar to Metro Exodus EE with an added screen space trace for further optimization.
Metro also performs screen space tracing first, so both are very similar. The difference seems to be in the denoisers, ray length settings, and stuff like that.

Also, the tracing itself seems to be extremely cheap in the game as software tracing loses just 1.8 ms compared to hardware tracing (8 ms for hardware tracing vs. 9.8 for software) on Ultra settings. Adding BVH for animated objects with the unobtanium reflection settings drops the software solution's performance considerably – 22 ms for hardware vs. 32 ms for software, maybe because their BVH builder is not nearly as fast for dynamic geometry as the driver's implementation, or maybe due to a much higher number of rays for reflections.
 
Status
Not open for further replies.
Back
Top