Digital Foundry Article Technical Discussion [2024]

[IQandHDR]

That's the thing. You worked in this field. I haven't. While you may immediately spot issues with a lighting system, I can sometimes not tell unless I have side-by-side comparisons. If baked lighting is good enough, for laymen, we won't notice what the ray tracing adds.

RT lighting stand out for me right away and I never worked in the industry, but I do have eyes.
Do not take me "hostage" in your argumentation, it is only valid for you, not everybody.

 

[VFX_Veteran]

I agree in principle, but I do think a lot of the time people get so tied up in the marketing checkmarks or narratives they want to drive that they aren't even looking at the images they are seeing critically. I'll give a few recent examples.

Hello Andrew! Of course they aren't. The only way they can see with a critical eye is if they've either seen and critiqued pipelines for years or sat in on technical directors critiquing pixels from a frame. Each of which is highly unlikely but we can't dismiss their opinions.

There's games that use largely-diffuse, pretty low density, probe-based "RTGI" (as if that term means anything anymore). On the plus side, doing dynamic traces against triangle geometry is a massive win from a production perspective vs. baking. On the other hand, the visual result is often only marginally better than one would expect from baked probes, with rampant leaking and no fine-grained detail at all (or sometimes only via screen traces). Often the RT scene used in these implementations is so simplified that you're basically just getting very broad, diffuse bounce with no fine grained contact or specular occlusion (which, IMO, are the main hallmarks of GI).

And this is what a lot of games are doing now, I agree. They are saying RTGI, but not knowing that it's RT against light probes (which were done offline). I think DF has been doing a good enough job though of being able to see what's going on with these games since they are getting experience critiquing them.

Obviously you have to do something to make games fit into 60fps console budgets or whatever, and I think overall the tradeoffs made in these implementations are reasonable. But these implementations (Indy w/o "full RT", The Finals, Avatar, Metro, etc.) should not really be discussed as if they are competing with implementations that are aiming for much higher fidelity, fine-grained, specular GI (various "full RT"/"PT" things, Lumen both modes, etc) because the results are not really comparable. Even if using the reviled SDF path, I would fully expect Lumen SWRT to look quite a bit better than the stock Indy RTGI, albeit at a higher cost.

Yup.

But that's the rub! The narrative often strays so far into "RT vs no RT" that people entirely lose the plot that there's a variety of solutions that sit in different places on the performance/quality spectrum. Even the critically important things like what does your tracing scene representation look like and at what frequency are you querying it and updating it are lost in the marketing and console wars. People start making arguments that are patently silly from a technical perspective because it supports whatever marketing narrative.

We can't expect the average gamer or reviewer to have that kind of in-depth knowledge though.

Yes RT is a foundational technology that will be used more as we move up the quality stack. This has been true for decades, with only the timelines being a question mark. I'm glad we're getting to the point where we can start to assume that some hardware support for ray queries is present in games we ship, and thus I'm really happy to see Indy take a dependency on that as it opens the way for more progress once that becomes the norm. But we don't need to pretend that "RT" in the title of a technique automatically makes it look great or better than something without it because in reality, "RT" is as much a marketing term now as a technical one. Let's look at the actual images and compare on the phenomena they model and at what accuracy and detail levels, algorithms aside.

Yea, but I blame the game companies for that though. Their engineers know what it means to RT and yet people on the receiving end is getting an overall "image" of what RT is becoming.

I used to laugh when companies talked about parallax occlusion mapping using ray tracing or RT a flashlight for some bounced light. Some of us have a lot of experience with understanding what a real ray-traced renderer is used for (i.e. the entire image instead of just a 2D ray-cast across a texture space). I, personally have gotten in heated arguments of what I experienced and what the gamers experience and no amount of trying to describe what "real" RT meant was going to help gamers understand what context I was trying to portray.

Let's not get silly, there's tons of polygonal edges in games, even Nanite ones. I don't really know what you mean by "lighting properly" because that's an arbitrary definition. What matters is how close the result is to a reference and I think Hellblade does relatively fine in that regard in the scope of its environments.

When I'm talking about "lighting" properly, I mean a raw path-traced algorithm that uses importance sampling with CDFs, BRDFs, and PDFs to represent light physics. I made that statement with the knowledge of my own experience over the years in offline rendering. So I may be extra picky with games. Ignore my hyperbole.

BTW, I think HB2 is a beautiful looking game.. it's the best usage of UE5 we have seen so far to date.

The texture point is one that I want to briefly mention though because it's one that I think really covers a lot of the amateur discussion of many of these games online. People in general respond to high resolution textures as if they are a primary thing that makes a game "good graphics". Obviously super smudgy stuff isn't visually appealing, but on the other hand really high res textures and normal maps on low res geometry that doesn't interact with detailed lighting is last gen technology.

Agree.

Dev hell and all that aside - just judging the outcome at launch - I don't think Stalker 2 is a particularly good looking game, despite fairly modern technology; I would not put it on my best of 2024 even as a runner up. I think people generally just think it looks good because it has pretty high texture resolutions and occasionally good geometric/asset density, but in terms of the lighting, it really does not strike me as anything special in the majority of what I've played and seen. Obviously people are free to like high resolution textures and such if they want to but I think from an visual and lighting level, it's not really in the same league as other things on the list.

Me either. That's one of the reason's I haven't bought it yet.

And to get on my personal soapbox again, I don't think stuff like the shadow quality of foliage in Stalker 2 or non-RT Indy, or to a slightly lesser extent Outlaws is acceptable in a 2024 game that gets put on a "best graphics of" list. (Obviously the ones that also have an RT path on PC get a pass assuming folks are speaking of those versions.) I don't really care if your textures and geometry are the size of pixels if your shadow map texels cover hundreds of pixels. I don't know why anyone else puts up with that either other than the fact that they got used to it. Honestly the biggest visual impact of the "full RT" paths often isn't even the GI parts, it's the fact that all lights get ray traced shadows.

You are on soapbox today huh? I think DF was judging these games based on all of the features the game could give -- basically how far can you drive your game in the graphics tech arena? In that case, I listened to the conversation with the PC hardware in mind. If they judged based on console iterations, then none of them are really doing anything over HB2 and wouldn't be worth the trouble. Nearly all of those games had to sacrifice an expensive main tech feature for the console. That's not DFs fault though.

 

[Below2D]

RT lighting stand out for me right away and I never worked in the industry, but I do have eyes.
Do not take me "hostage" in your argumentation, it is only valid for you, not everybody.

I wasn’t addressing you and I bet I can post screenshots with RT on and off and you wouldn’t be able to tell the difference. Sometimes, baked is damn near identical to rt. Other times, it’s obvious. Bad geometry and low polygon details are always obvious.

 

[Remij]

Dev hell and all that aside - just judging the outcome at launch - I don't think Stalker 2 is a particularly good looking game, despite fairly modern technology; I would not put it on my best of 2024 even as a runner up. I think people generally just think it looks good because it has pretty high texture resolutions and occasionally good geometric/asset density, but in terms of the lighting, it really does not strike me as anything special in the majority of what I've played and seen. Obviously people are free to like high resolution textures and such if they want to but I think from an visual and lighting level, it's not really in the same league as other things on the list.

I'm curious as to what your top 3 games would be for this year.

For my own personal list, I think it's:

1. Silent Hill 2 - Everything in that game is beautifully done, especially on PC maxed out.
2. Hellblade 2 - Some truly beautiful environments and character models and animations.
3. Black Myth Wukong - Stunning environment detail and lighting.

Runner ups - Path of Exile 2, Warhammer 40K: Space Marine 2, and Astro-Bot.

 

[Andrew Lauritzen]

The only way they can see with a critical eye is if they've either seen and critiqued pipelines for years or sat in on technical directors critiquing pixels from a frame. Each of which is highly unlikely but we can't dismiss their opinions.

For sure, and I try to qualify everything I say from a technical perspective with the fact that in the end, this is personal preference and to a large extent games look great because of great artists more than great tech. But obviously the two together is where you get truly exceptional stuff.

We can't expect the average gamer or reviewer to have that kind of in-depth knowledge though.

True, but I think it's fair to have the more technical discussion at B3D specifically. It's not like I would go to other random public forums and expect to chat in depth about any of this with gamers.

I used to laugh when companies talked about parallax occlusion mapping using ray tracing or RT a flashlight for some bounced light. Some of us have a lot of experience with understanding what a real ray-traced renderer is used for (i.e. the entire image instead of just a 2D ray-cast across a texture space). I, personally have gotten in heated arguments of what I experienced and what the gamers experience and no amount of trying to describe what "real" RT meant was going to help gamers understand what context I was trying to portray.

I'm actually more upset about the fact that "path tracing" means nothing now than raytracing, because while I agree it has a more specific meaning in offline rendering, it has always been a fairly generic term in real-time, applying to anything that intersects a ray with [some sort of geometry representation]. Even if you restrict it to "just triangle RT", there's a disconnect in real-time because the contents of the structure being intersected against is rarely the full fidelity geometry, but no one is in a rush to advertise what theirs looks like in practice

Anyways I don't mean to gate keep the terminology. It's more just that I think it is increasingly confusing enough that people need to spell out much more specifically the details before they declare something technically inferior/superior. Saying "this thing is using hardware raytracing and this other one isn't" increasingly doesn't really mean much in terms of what to expect from the quality perspective.

When I'm talking about "lighting" properly, I mean a raw path-traced algorithm that uses importance sampling with CDFs, BRDFs, and PDFs to represent light physics. I made that statement with the knowledge of my own experience over the years in offline rendering. So I may be extra picky with games. Ignore my hyperbole.

Nah, it's good. Offline path tracing is the gold standard, and it's a great thing to compare to. From that perspective though even the stuff people have called "path tracing" in games is not really doing all of that yet. Things are getting there piece by piece but real-time is often an exercise in figuring out which parts of it are really important vs. which parts are more edge cases (which can often be content-specific).

Aside: I would love it if we could get better reference images to compare to in these games/reviews. Often when people are showing off something in a screenshot or video I am fairly sure there are some pretty big errors being overlooked vs. what a reference image would look like; in the odd case people will often declare a shot that is further from what I would expect the reference image to be to be the superior one too. Obviously matching reference is not always the goal and there are other considerations, but it's a useful tool. While there would be ways to grab/dump the geometry from a BVH in a game fairly similarly and trace against that, I know from even the complexity of the reference path tracer in Unreal that there's so many other things that go into a frame than just the ray intersections and materials that make it look quite different and thus hard to compare. Alas, would love it if it were something we were able to see more frequently, like if it were an option in some of the game's photo modes or similar.

You are on soapbox today huh?

The benefit/curse of actually having some holiday time to finally get to play some of these more recent games

I think DF was judging these games based on all of the features the game could give -- basically how far can you drive your game in the graphics tech arena? In that case, I listened to the conversation with the PC hardware in mind.

Yeah that is my assumption as well, although they do waffle a bit between comparing performance as well, in which case obviously the PC "full RT" paths are in a completely separate tier. There's no one way to compare these games and it's fine for everyone to have their own biases as long as we keep the conversation respectful. If I were to make a personal list I'd almost want to split it between games I find technically impressing from a rendering perspective in a quality/perf ratio sort of way (the combination of the two is really what is difficult), and games that just look great because they have great art and use even more focused/limited tech well to that end.

 

[trinibwoy]

I don't get why John puts Indy higher than Hellblade 2 for scalability. Indy needs a ton of VRAM and if you don't have that, it's going to be unplayable. This alone makes it a game that scales terribly. There's a lot of 6 GB and even 4 GB cards that can't run Indy but can run HB2 flawlessly.

Also Indy is not a good overall graphics package, because the characters look terrible. And the asset quality is not that great compared to Nanite driven titles.

Yeah I wouldn’t give Indy top billing. The lighting tech is great but lots of other stuff isn’t top tier.

 

[cheapchips]

Non-Nanite geometry would be mesh shaders, which can go just as hard as Nanite if Alan Wake 2 is anything to go by.

and Indy is no where near Alan Wake 2's geometry level.

I've only played the first wood area and town of AW2. It didn't feel as excitingly rich as some areas of Indy do.

 

[Cyan]

There's games that use largely-diffuse, pretty low density, probe-based "RTGI" (as if that term means anything anymore). On the plus side, doing dynamic traces against triangle geometry is a massive win from a production perspective vs. baking.

On the RT front I don't think the current tech landscape is actually that complicated. We want dynamic GI across the stack because the production burden of baking is increasingly unworkable with modern amounts of content. Therefore stuff like real-time probe diffuse GI makes sense even if it doesn't necessarily look much better than offline baking (which guess what guys... has always been raytracing...) would have.

do you mean that if developers start using/requiring Raytracing by default on all games they will save money?

In otder to save money, wouldn't be the best idea to use the most powerful machine to make actual full path traced offline graphics and then converting those assets into baked lighting? Just curious.... And sorry if I didn't understand you correctly.

 

[Below2D]

You must not be playing Indy on a high-end PC because the majority of your complaints are from the console.

I played at max settings on a 4090+13900K. Alex also touched upon the LOD issue in the video. It occurs even at max settings and you have to manually tweak the config files (or was it a command?) to increase it.

 

[Cyan]

Best graphics of 2024. Graphics of the year award.

this has been a great year for gaming in overall quality. I guess DF chose Indy not because it is the true top at graphics quality alone, but maybe because it is the best game mixing graphics fidelity and performance. Just my guess... Other than that, as lovely as Indy lighting is even on non path traced graphics, many of the games in that list are as good if not better in some aspects imho.

 

[trinibwoy]

do you mean that if developers start using/requiring Raytracing by default on all games they will save money?

In otder to save money, wouldn't be the best idea to use the most powerful machine to make actual full path traced offline graphics and then converting those assets into baked lighting? Just curious.... And sorry if I didn't understand you correctly.

Baking takes time and time is money. You need to redo the bake whenever you change anything in the scene. Not to mention baking doesn’t help at all with dynamic assets. It’s a dead end.

 

[Scott_Arm]

Non-Nanite geometry would be mesh shaders, which can go just as hard as Nanite if Alan Wake 2 is anything to go by.

and Indy is no where near Alan Wake 2's geometry level.

Mesh shaders is just an api. It’s not Nanite vs Mesh Shaders (It even uses them). It’s Nanite vs other geometry pipelines that may use mesh shaders. I find Mesh Shaders gets referenced a lot as if it’s a full solution in itself, but it’s not.

Alan Wake 2 has amazing visuals and a lot of geometric detail, but credit where credit is due to Remedy for creating an engine system that’s capable of handling it.

 

[Scott_Arm]

@Andrew Lauritzen @Remij

I like the calllouts for Path of Exile 2. It’s a really surprisingly good looking game and it runs very well. I’m sure if I zoomed in the camera I could see some limitations to the models and textures, but it works very well. I wish the particle effects contributed to lighting the scenes(I should check if it does on ultra GI but I doubt it).

Game can already be a lot visually when spells are going off, but it would be a nice option. The animation, particle effects and overall art direction are pretty strong.

Kind of surprised the game isn’t getting more attention outside the arpg community.

 

[IQandHDR]

I wasn’t addressing you and I bet I can post screenshots with RT on and off and you wouldn’t be able to tell the difference. Sometimes, baked is damn near identical to rt. Other times, it’s obvious. Bad geometry and low polygon details are always obvious.

A lot of people can see the differences, the main "argument" from people against RT is not that they cannot see the image qulity difference, it is that they do not think the performance hit is worth it.
You do not need to work in the industry to see eg. "light bleed".

 

[davis.anthony]

They do not think the performance hit is worth it.

They said the same thing 20 years ago when HDR lighting started to be introduced.

The only difference is back then the gen on gen GPU performance uplift was enormous, so it only took a few GPU generations to make it a 'turn on and forget' setting.

 

[IQandHDR]

They said the same thing 20 years ago when HDR lighting started to be introduced.

The only difference is back then the gen on gen GPU performance uplift was enormous, so it only took a few GPU generations to make it a 'turn on and forget' setting.

Same with RT.
Look at the generational uplift.

People just assmue today that their entry level GPU can do every at max settings.
People have bad memories, this is "The Witcher 3" at launch:

The Witcher 3 Video Card Benchmark - Poor Software Optimization by CDPR | GamersNexus

stub Benchmarking the Witcher 3 proved to be more cumbersome than any game we've ever benchmarked. CD Projekt Red's game doesn't front the tremendously overwhelming assortment of options that GTA V does – all of which we tested, by the way – but it was still a time-consuming piece of software to...

gamersnexus.net

Only thing that have changed since 2015 is that the whining minority has becaome far to vocal for their numbers and they forgot about the past.

Was no better in 2004 either:

Matching Doom3 With The Best Graphics Card

Which graphics cards offer maximum-quality Doom3 play? We benchmarked a selection of newer and older models from ATI and NVIDIA to find out.

www.tomshardware.com

 

[Below2D]

A lot of people can see the differences, the main "argument" from people against RT is not that they cannot see the image qulity difference,

Again, you're missing the point. The point is that there are levels to how accurate baked lighting is versus ray-traced lighting. Sometimes, the differences are subtle such as with this case.

Whereas at other times, the difference looks literally almost like night and day such as with this case:

it is that they do not think the performance hit is worth it.

No, sometimes, the differences are minor unless you look for them. Low-quality geometry cannot be hidden OR FAKED. Anyone can spot its ugly head whenever it appears. It's possible to have good quality lighting even without ray tracing. Some people are good enough to approximate it and give excellent results.

You do not need to work in the industry to see eg. "light bleed".

That's assuming there is light bleed occurring in the scene, which isn't always the case.

 

[troyan]

Non-Nanite geometry would be mesh shaders, which can go just as hard as Nanite if Alan Wake 2 is anything to go by.

and Indy is no where near Alan Wake 2's geometry level.

DigitalFoundry's video is about games released in 2024. Which other non UE5 game pushes as much geometry as Indiana Jones and was released this year?

This is an example of pushing the limits with tradional rasterizing:

 

[Liandry]

My top 3 graphics this year.
Third place Indiana Jones and the Great Circle. I don't know how anyone can say there is not much geometry. There is a LOT of geometry, more than in average game. High res textures, a lot of bump mapping. There is a mix of geometry and bump maping and relief of objects looks very realistic. Photorealistic lighting, very good realisation. RTGI. And that game runs in 60 fps on both Xbox consoles. Great achievement, especially for XSS.
Secons place Stalker 2. Amount of geometry is just super high. I now Play Far Cry 6 and that games also looks very good, but ist's clear that it is last gen game. Difference in amount of geometry is unbelivably high, and that difference is everywhere. Even each brick in walls is separate object! Texture resolution is very high. Different materials. Amazing lighting, especially when there is storm. Super high detailed landscape and ground. And open world, realtime day and night cycle and randon change of weather.
First place Senua's Saga: Hellblade 2. That game is absolutely best this year in terms of graphics. There is no game for now what is even close. Amount of geometry os just beyond everything. Characters are most detailed and realistic. Many times I turned on photomode during cutscenes and looked on characters from different angles or with zoom. Sometimes looks like CG. Textures not the best but good enough. Lighting is top. But there is also one important moment. Then you complete some episode and camera moves ahead high above ground for some kilometers ahead, there is not a singlew pop-up. And also this game have truly amazing particles, maybe best particles in any game for now.

 

[Dampf]

Mesh shaders is just an api. It’s not Nanite vs Mesh Shaders (It even uses them). It’s Nanite vs other geometry pipelines that may use mesh shaders. I find Mesh Shaders gets referenced a lot as if it’s a full solution in itself, but it’s not.

Nanite doesn't use Mesh Shading in any way that actually matters, because otherwise the 2060 Super would pull far ahead of the 5700XT just like we have seen in Alan Wake 2.