Digital Foundry Article Technical Discussion [2023]

[mr magoo]

Yea, it's pretty bad optics. As they said in the video there, they've had a year since the game launched on PS5 and PC to improve things and they should have capitalized on this new release to fix longstanding issues with the game, and hell, throw their loyal console gamers a bone and maybe give them something exclusive to stand out from the others.

But no.. this just continues the narrative that Xbox doesn't manage studios properly, and that their hardware is continuing to underperform relative to their competitor.. and that they don't think enough of their core fanbase to do something special for them.

They should be firing on all cylinders by now in the gen.. and yet it still looks like nothing is changing. They really need some wins right now.. because honestly looking at my twitter, their hardcore fanbase is starting to ask hard questions.. and that's not good.

Everything you say is true and pretty much sums up everybody thoughts on current state of things. But let me play devils advocate for a second perhaps High five rush had something to do with this? Perhaps they were focused on this title which resulted in worse quality of Ghost Wire.
But you are right in general I have thoughts about selling my XSX and buying pc parts instead. Things are bad, I have my XSX for almost 3 years now and the only thing that keeps me using it is GP.
If I wouldn’t have to upgrade my whole PC I would be selling XSX right now.

 

[RobertR1]

For path tracing, is there a point of diminishing returns for the number of light bounces?

 

[davis.anthony]

For path tracing, is there a point of diminishing returns for the number of light bounces?

Impossible to say at the moment, playing with ray counts and ray bounces in CP2077 Overdrive mode does show diminishing returns after 4 bounces but it's a game with mainly last generation geometry and assests.

A more geometrically complex next generation game could benefit from more bounces as there will be more geometry and objects to bounce around that may potentially benefit from more bounces.

 

[Inuhanyou]

For path tracing, is there a point of diminishing returns for the number of light bounces?

Based on Alexs previous statements 4 to 6 are the limit. Any more than that doesn't give much improvement compared to how much they cost

 

[mr magoo]

https://twitter.com/x/status/1648325508185325568

 

[GuardHei]

I kinda feel the ssao implementation is improved in Tears of the Kingdom. In BOTW it's basically thins outline around object edges, but in TOTK I somehow feel the distance is extended and looks more close to HBAO.
I'm personally also curious about the new GI system. The old one seems to be a single gi probe updating with the camera position, and only captures important and (mostly) static landscapes, so you kinda have serious light leaks inside tunnels or underground (cuz these are surronded by non-landscape meshes), which doesn't seem to be the case in TOTK? Wonder whether they change the method or just capture more stuffs.
Otherwise the tech seems to be the same. I never understood why DF said the cloud was volumetric earlier. To me it looks like billboard with some hacky shading technique to create the volumetric effect (even it is volumetric, switch is prolly ok to handle it as in xb1de/2/3), maybe some nice normal map trick to create the illution? Screen space reflection is back. Again not in the over-world, but you can see it present in the puzzle sections (the rusty metal floor does have a SSR). Far shadows still look pre-baked and interpolated between times. The only effect not seen yet is raymarched godrays.

 

[jlippo]

Impossible to say at the moment, playing with ray counts and ray bounces in CP2077 Overdrive mode does show diminishing returns after 4 bounces but it's a game with mainly last generation geometry and assests.

A more geometrically complex next generation game could benefit from more bounces as there will be more geometry and objects to bounce around that may potentially benefit from more bounces.

Yup, future games will be interesting.

With proper refraction the requirements will rise.
10 bounces will not be enough in some cases.

 

[GuardHei]

How does Metro Exodus Enhanced handle the secondary bounces? I vaguely remember it was DDGI in the original version, not sure about the enhanced one.
Personally I don’t know about the cost of managing a whole additional grid of probes, but then it also provides infinite bounces

 

[mr magoo]

How does Metro Exodus Enhanced handle the secondary bounces? I vaguely remember it was DDGI in the original version, not sure about the enhanced one.
Personally I don’t know about the cost of managing a whole additional grid of probes, but then it also provides infinite bounces

Metro Exodus Enhanced Edition - the ray tracing showcase tested on Xbox Series X/S | Eurogamer.net

"Light from the sun, the moon - or objects that emit light - taps into RT, illuminating any given scene with accurately calculated light bounce. The original Metro Exodus's PC RT solution shipped with just one light bounce, while the Enhanced Edition - even on consoles - works with infinite bounces, calculated over time. As a consequence of this, usual screen-space ambient occlusion (SSAO) isn't required. It's all part of the global GI solution delivered by ray tracing. It's also far easier to show rather than tell, so please check out the video embedded below to appreciate how this technology works."

 

[GuardHei]

Metro Exodus Enhanced Edition - the ray tracing showcase tested on Xbox Series X/S | Eurogamer.net

"Light from the sun, the moon - or objects that emit light - taps into RT, illuminating any given scene with accurately calculated light bounce. The original Metro Exodus's PC RT solution shipped with just one light bounce, while the Enhanced Edition - even on consoles - works with infinite bounces, calculated over time. As a consequence of this, usual screen-space ambient occlusion (SSAO) isn't required. It's all part of the global GI solution delivered by ray tracing. It's also far easier to show rather than tell, so please check out the video embedded below to appreciate how this technology works."

Oh I see. So the original version is just one bounce, and it is the enhanced version which uses ddgi

 

[Shifty Geezer]

For path tracing, is there a point of diminishing returns for the number of light bounces?

Yes, but it depends on context. Filling an opaque room (Cornell Box) with bounce light, I think 4 bounces generally does it. Throw in some reflections and refractions and you could need dozens to fully resolve the light paths. The issue is how much difference the next iteration makes relative to the current accumulate light value. If the fifth bounce is only 1% of the current surface illumination, no-one will notice on a painted wall, but that fifth bounce could be the difference between a lit sample and a blank sample, 100%, if the path is complex.

The other parameter is sample count. Fewer bounces of more rays will give better shadow definition.

 

[hughJ]

@Dampf Yah, the differences between the path traced and the raster are pretty big in most of the pictures and video I've seen. Path tracing is way more natural looking, as it should be. All path traced games really need is virtual objects that you can't see in-game that bounce, block or diffuse light like they do in film. They basically have these canvas or boards that can bounce light in a variety of warmths, or block it if they're black. That way you can add or remove lighting in a scene that looks very natural, and it's happening off screen so you can't really tell that things are being manipulated. I've worked with a friend who is a photographer and spent a bunch of time holding those things on small shoots.

I'm not sure how path-traced equivalents of invisible bounce boards would work in a dynamic environment. It's fine for film where you have control over the camera frame and scene objects, but in the context of a video game you don't. The phantom light emitter/reflector would cause dynamic objects getting near it to get mysteriously bright (and abruptly dark again if it should pass through it), and whatever light that object is receiving would mean an equivalent amount of light not getting projected onto the scene elements that you were trying to dress-up, so you'd get unmotivated/unsourced shadows moving around. I think if you were dead set on cheating with PT you'd be better off adding an entity into the level where any object within a given radius gets a multiplier, or better yet, just do an iris/exposure and color correction buffer effect.

 

[trinibwoy]

Yeah that stuff sounds real flaky as long as the user has control of the camera. If an artist needs explicit control over lighting in a scene maybe they can get there by tweaking light and material parameters. Faking lighting like in movies will be pretty hard in games whether you're rasterizing or path tracing. You're probably better off just baking the lighting at that point.

You will call me crazy but maybe I do like more the holywood look of the raster version.
But I can see the shadows and background light source are better in the RT.
Can we have a mix? The more raster tone with the details of RT?

Are you referring to the yellowish tone of the raster screenshot? I prefer it too but I doubt it has anything to do with raster vs path tracing tech. It's just the difference in the light sources used in each renderer. Rasterization doesn't make things turn yellow

 

[Silent_Buddha]

I'm not sure how path-traced equivalents of invisible bounce boards would work in a dynamic environment. It's fine for film where you have control over the camera frame and scene objects, but in the context of a video game you don't. The phantom light emitter/reflector would cause dynamic objects getting near it to get mysteriously bright (and abruptly dark again if it should pass through it), and whatever light that object is receiving would mean an equivalent amount of light not getting projected onto the scene elements that you were trying to dress-up, so you'd get unmotivated/unsourced shadows moving around. I think if you were dead set on cheating with PT you'd be better off adding an entity into the level where any object within a given radius gets a multiplier, or better yet, just do an iris/exposure and color correction buffer effect.

In theory the artist should have control over how many bounces a light is allowed to have, the distance a light is "allowed to travel", the "volume" a light is allowed to exist in, special surfaces that can only be lit by that light source (for example a light meant to highlight a character's face would be problematic if something partially occulded the character's face and the light lit up part of that occluding object), etc.

Those controls should allow you to emulate many lighting tricks used by light riggers. So, this should potentially remove the need of say, a secondary light used specifically and only to remove shadows cast by a light that is meant to highlight a character's face or even backlight a character strongly for mood purposes.

Unlike film the light source or sources can still be offscreen and follow the character.

It is certainly interesting to think that one unintended side effect of non-RT lighting is that you get many of these artistic uses of lighting "for free". Lighting artists for games are going to have to learn how to use RT lighting to achieve the same effects with a dynamic scene where the player has control of the camera.

Regards,
SB

 

[DavidGraham]

I'm not sure how path-traced equivalents of invisible bounce boards would work in a dynamic environment.

If you really want to, you can program a hidden rasterized point light and use it as a hero light in any cutscene you want. Lumen and UE5 does something like this just fine, in the Matrix demo, they used an area light for cutscenes, with hardware RT shadows on the characters during the cut scene. It worked fine.

 

[Flappy Pannus]

For PS5 SX, res numbers:



The graphic denotes dynamic res, but Tom speaks like it's fixed, never really mentions a range, so who knows. Just the one mode, but it's a very stable 60fps.

PS4/Xbox are a relatively smooth 30fps at 1080p/900p respective. PS4 Pro/One X both render at 1400p with 30fps, and hold it well.

 

[Inuhanyou]

For PS5 SX, res numbers:

View attachment 8789

The graphic denotes dynamic res, but Tom speaks like it's fixed, never really mentions a range, so who knows. Just the one mode, but it's a very stable 60fps.

PS4/Xbox are a relatively smooth 30fps at 1080p/900p respective. PS4 Pro/One X both render at 1400p with 30fps, and hold it well.

It's cuz he never saw any drops in res even if it might be dynamic

 

[scently]

The shadows are sharper on the PS5 vs the SX.

 

[Flappy Pannus]

The shadows are sharper on the PS5 vs the SX.

The shadows on PC Ultra more closely resemble the SX from other videos on the PC that I've seen. Could be a difference in soft shadow falloff.

Same resolution but higher cascade settings on PS5 vs PC and XSX.

Is there a video comparing the console versions to the PC with equivalent settings comparisons?

 

[Globalisateur]

The shadows are sharper on the PS5 vs the SX.

Same resolution but higher cascade settings on PS5 vs PC and XSX.