Digital Foundry Article Technical Discussion [2024]

[iroboto]

How much better or how noticeable multiple light bounces are in a game is subjective to scene and art. A different view is the practical needs of devs, you can see this hour long playthrough of Avowed just how much lighting artists need to "work around" UE5's limited light bounces, literally needing to come up with multiple types of new light sources just to get interiors viewable at all despite multiple places in the level letting in sky and sunlight. With say, 4 light bounces instead of 2 w/feedback there'd be less need to place to extra light sources all over.

I would say they are unlearning what they are used to designing, and now learning how to light appropriately. It’s challenging because even in movies and photography, most lighting is augmented to obtain the look you want.

I didn’t see anything in particular there that was odd. Resident evil and many games always seem to have abandon/hidden areas that are for some reason pre-lit by candles. Maybe the going forward is to do what all table top RPGs do; give the players a torch.

 

[Frenetic Pony]

I would say they are unlearning what they are used to designing, and now learning how to light appropriately. It’s challenging because even in movies and photography, most lighting is augmented to obtain the look you want.

I didn’t see anything in particular there that was odd. Resident evil and many games always seem to have abandon/hidden areas that are for some reason pre-lit by candles. Maybe the going forward is to do what all table top RPGs do; give the players a torch.

That's, not what I'm talking about. Games have had prebaked GI for almost 20 years now, they all "light properly", and it is time consuming. Realtime GI ideally cuts out bake times to speed up artist workflow. But the practical problem with UE's current realtime GI is that you can't count on strong light sources bouncing around, and so can't recreate natural environments where the player can see at all unless you have a hole in the roof every 10 meters. If a room doesn't have a giant hole you need to come up with a glowing mushroom or a glowing crystal or worse you need to put in a fake light, even if you don't want to and that's not the art direction you're going for, in order for the player to even see anything. Then you need to place them decently, everywhere, every single room, and move it around every single time you move the rest of the room around.

If there were a decent number of light bounces this timestamped hallway wouldn't need this glowing mushroom, but because there's limited light bounces someone had to make this asset, then someone had to put this asset in the game, then every revision or iteration it has to be thought about or moved around or etc. All of this could be solved by an extra bounce or two, freeing up both art direction and developer time:

 

[Dictator]

That's something that bugs me about this. It's a big problem in wukong. Some parts of the maps are just pitch black, I really don't like how it looks.

Did wukong really need to use lumen? It's a (mostly) linear game with no time of day transitions. Naughty dog games have really good bounce lightning with no visibility problems, with dynamic objects being the only negative.

I'd say that baked lighting still has it's place for those types of games.

By the way, isn't lumen supposed to be infinite bounces? Or are black spots causes by something else?

PS: I just learned that avowed will have a day to night cycle, so what I said doesn't apply.

You must not be noticing how often Wukong changes the sun position. If you play the first level for example, you can see the sun position changes with each corridor you go through, sometimes you can even see it occuring If you are attentive.

 

[Charlietus]

You must not be noticing how often Wukong changes the sun position. If you play the first level for example, you can see the sun position changes with each corridor you go through, sometimes you can even see it occuring If you are attentive.

Even if the sun changes position, maybe a similar outcome could have been achieved with baked lighting.

I wonder if it would be possible to ship a console version with baked lighting while the PC version has lumen-full ray tracing. I don't think that's ever going to happen just because of the additional workload on the developers.

We have just one good example of a game with the UE5 features enabled on consoles in Fortnite, literally all other games are at least a bit problematic.
Some developers are being a little bit too eager to enable those features without thinking about the consequences.

 

[RobertR1]

Eh let these consoles age poorly so we can accelerate a new gen. These things weren’t great to begin with so sooner this gen is over the better.

 

[Nesh]

Eh let these consoles age poorly so we can accelerate a new gen. These things weren’t great to begin with so sooner this gen is over the better.

I feel that these consoles have barely taken advantage of and there is still a lot of life in them

 

[Charlietus]

Eh let these consoles age poorly so we can accelerate a new gen. These things weren’t great to begin with so sooner this gen is over the better.

Lol, reluctantly agree

 

[trinibwoy]

By the way, isn't lumen supposed to be infinite bounces? Or are black spots causes by something else?

Lumen does support infinite bounces so it may be an issue with the asset setup. If there are black spots it could be due to missing coverage from Lumen’s surface cache which doesn’t work that great with complex meshes.

Lumen GI is an incredible piece of software engineering but it’s a bit overengineered and therefore fragile. The short term benefit is that it runs on everything today but I don’t expect it will age well without significant enhancements. The multiple levels of low resolution screen, surface and world space caches, slow update rates and sensitivity to asset layout just seems unnecessarily complicated when compared to something more streamlined like ReSTIR

 

[RobertR1]

I feel that these consoles have barely taken advantage of and there is still a lot of life in them

It’s zen2 and rdna2. There’s no magic here. It’s all well known and explored.

I do look forward to people paying twice for a small performance bump soon though and overselling it to themselves.

Edit: imagine if there is no pro. Then we are likely looking at a 2026 announcement of a new console which will be many times over faster and more capable. Now we will get a stretched generation well into 2028 before seeing something new.

I’m a tech fan and want to see things moving forward.

 

[Nesh]

It’s zen2 and rdna2. There’s no magic here. It’s all well known and explored.

I do look forward to people paying twice for a small performance bump soon though and overselling it to themselves.

Edit: imagine if there is no pro. Then we are likely looking at a 2026 announcement of a new console which will be many times over faster and more capable. Now we will get a stretched generation well into 2028 before seeing something new.

I’m a tech fan and want to see things moving forward.

To be honest seeing how technology is moving and how the cost effectiveness has been for the past 7 years, I don't see much leap yet to warrant a sooner next gen launch of consoles.
Its going to be boring. Its gonna look more like "now the same games with true 4k and 60fps". Which is pretty much what this gen sounded like.
There also seems to be a stagnation in game releases that people want to play.

 

[RobertR1]

To be honest seeing how technology is moving and how the cost effectiveness has been for the past 7 years, I don't see much leap yet to warrant a sooner next gen launch of consoles.
Its going to be boring. Its gonna look more like "now the same games with true 4k and 60fps". Which is pretty much what this gen sounded like.
There also seems to be a stagnation in game releases that people want to play.

There’s more to it than graphics. A major jump in AI would lead to much better and dynamic interaction with NPC’s. Imagine a RPG where the NPC’s are adapting their dialogue based on events using natural language processing and not limited to a script tree.

 

[see colon]

There’s more to it than graphics. A major jump in AI would lead to much better and dynamic interaction with NPC’s. Imagine a RPG where the NPC’s are adapting their dialogue based on events using natural language processing and not limited to a script tree.

We could have this today using "the power of the cloud®", and a fallback to more traditional interactions for offline play. I think there are a couple of things holding this back, though. One being the unpredictability of having such interactions with an NPC, who may go off script and lead the player on a tangential conversation that is so far off script it breaks the 4th wall, hints at non-existent quests of features, or veers into some politically sensitive topic that the developers never intended. The second is, if you want that character to be voiced, you would have to use an AI voice model to take into account the freeform conversation. Then you get into the can of worms of replacing voice actors with AI voice actors, which itself can be a sensitive subject.

 

[Nesh]

There’s more to it than graphics. A major jump in AI would lead to much better and dynamic interaction with NPC’s. Imagine a RPG where the NPC’s are adapting their dialogue based on events using natural language processing and not limited to a script tree.

We are going to need more time for that. This is not the right time. Also people will require a lot more than NPC interactions

 

[raytracingfan]

How much better or how noticeable multiple light bounces are in a game is subjective to scene and art. A different view is the practical needs of devs, you can see this hour long playthrough of Avowed just how much lighting artists need to "work around" UE5's limited light bounces, literally needing to come up with multiple types of new light sources just to get interiors viewable at all despite multiple places in the level letting in sky and sunlight. With say, 4 light bounces instead of 2 w/feedback there'd be less need to place to extra light sources all over.

I'm curious how performance and visuals of 4 bounces would compare against 1-2 bounces and sampling from DDGI, NRC, or SHaRC afterwards. 1 bounce + DDGI worked well enough for Metro Exodus EE.

 

[Andrew Lauritzen]

We have just one good example of a game with the UE5 features enabled on consoles in Fortnite, literally all other games are at least a bit problematic.
Some developers are being a little bit too eager to enable those features without thinking about the consequences.

I mean hellblade 2 is clearly a good example right? Admittedly 30fps, but the issues with it are all pretty minor and less than the average game release.

 

[Andrew Lauritzen]

The multiple levels of low resolution screen, surface and world space caches, slow update rates and sensitivity to asset layout just seems unnecessarily complicated when compared to something more streamlined like ReSTIR

But this is the rub, and has been forever. You can write a path tracer in a couple days as an undergrad and it's simple and elegant but the more you want to make it fast the more complicated it becomes. You always have to find a balance of maintainability/complexity and performance and the reality is if you want to fit on current gen consoles (or even midrange PCs), stuff like ReSTIR DI is just not even in the discussion. That said, I think there are many interesting algorithms to come in the middle area here, and I don't think RTXDI as it is today represents any sort of end state (nor would I imagine NVIDIA would argue it does). I suspect people can get similar results for cheaper in the future as well, but time will tell

All of these techniques are still very much under active development and all the comments about marketing names apply on both sides. As much as "RTX" or "full RT" doesn't really mean anything in terms of technology, neither does "Lumen"; it's just a marketing bucket under which a bunch of algorithms are grouped, but those algorithms and their details change frequently as needed.

 

[Charlietus]

I mean hellblade 2 is clearly a good example right? Admittedly 30fps, but the issues with it are all pretty minor and less than the average game release.

Oh yeah, i forgot about hellblade 2 yesterday.

 

[RobertR1]

I mean hellblade 2 is clearly a good example right? Admittedly 30fps, but the issues with it are all pretty minor and less than the average game release.

It's more a walking simulator more than a game. Very scripted and linear. I'd imagine tuning that is a lot simpler than something a lot more dynamic in it's pathing, weapons, options, actions and such.

 

[trinibwoy]

But this is the rub, and has been forever. You can write a path tracer in a couple days as an undergrad and it's simple and elegant but the more you want to make it fast the more complicated it becomes. You always have to find a balance of maintainability/complexity and performance and the reality is if you want to fit on current gen consoles (or even midrange PCs), stuff like ReSTIR DI is just not even in the discussion. That said, I think there are many interesting algorithms to come in the middle area here, and I don't think RTXDI as it is today represents any sort of end state (nor would I imagine NVIDIA would argue it does). I suspect people can get similar results for cheaper in the future as well, but time will tell

All of these techniques are still very much under active development and all the comments about marketing names apply on both sides. As much as "RTX" or "full RT" doesn't really mean anything in terms of technology, neither does "Lumen"; it's just a marketing bucket under which a bunch of algorithms are grouped, but those algorithms and their details change frequently as needed.

For sure and there’s also quality and compatibility to take into consideration. Lumen is probably the right solution for the current generation of hardware and I’m sure it will evolve as better hardware comes online and other techniques become more viable.

When we talk about similar results it’s always in context of games designed around some lowest common baseline. Baked lighting is the perfect example. If you design your game to have static elements baked lighting is awesome. It’s when the goalposts move that things actually get interesting.

 

[DavidGraham]

DF is impressed with Astro Bot.