Unreal Lumen (Global Illumination without RT)

I dont think they are working on realtime GI right now. DDGI was dropped too.

 

Maybe it wasn't dropped. Maybe RTXGI will come next in 2023 and have a fallback to baked APV light probes.

 

Because that is how Unity has and most likely will continue to operate. Every demo that they have released in the past had custom functionality built for it that never made it but was vital for the demo to function.

 

What to do, should I go Unreal...

 

Yes

 

Absolutely. Go there then return. Basically what everyone does once C++ does a spinning kick with their undocumented API.

 

I wonder if Unity has a plan for when they will lose the documentation and tutorials advantage as soon as they push DOTS on everyone.

(who am I kidding, they don't have a plan, they're more interested in edgy twitter posts and making sports car analogies instead of making plans for DOTS and proper docs respectively)

 

Of course they have a plan. But the question of the hour is will anyone read it?

 

An open source game engine written entirely in like, Rust, would be the dream. Good documentation, memory safe, heck build a job system in and force all multi-threading to go through that, thread safe as well.

That's what Embark studios is doing, but it doesn't seem like their stuff is going to be available beyond some demos on github anytime soon.

 

I love what Rust is doing but it's the very early days still. As the codebase grows to engine work things will slow down quite a bit (in progress).

 

SSGI in Unity works better than I expected.
However, I thought that the part where the closed space did not darken needed improvement.
(Scene with directional lights or physical sky)
Placing a reflex probe will improve it a bit, but it will require more manual work.
AVP seems to be automatic and convenient, but I wasn't happy with the current results.
If the building is not along the axis of the AVP grid There is a fairly unnatural light and dark area.

Lumen will be properly darkened in a closed room.
Also, there is no need to manually place the probe or bake anything in advance.

I hope the entire scene of the enemy demo will be shared.
(Heretic was a great demo, but only the characters are still shared.)
If the Enemy demo can experience real-time GI with the push of a play button,
many would expect Unity.
The ancient valley demo of UE5 really worked with the push of a play button.
(Even my Geforce 1060. Of course, the frame dropped.)

 

Unity could have very easily had all of this from the very beginning. A different language may have come with improvements through forcing certain aspects in place but things like good documentation are not automatic.

 

Please keep on topic.

 

Have you checked this out by them? It's not a complete game engine but the global illumination is state of the art. And I'm sure with enough programming knowledge can be incorporated into rust game engines like bevy
https://github.com/EmbarkStudios/kajiya
https://rustrepo.com/repo/Seabass247-bevy-kajiya
https://github.com/Seabass247/bevy-kajiya-playground

 

I only hope Unity comes up with something before they deprecate Enlighten again. Or they could upgrade to recent SDK version of Enlighten (from what i've read they improved lightmaps build times etc).

 

HDRP has various GI things, what is it you're looking for from GI?

 

It could be a bad news for us (their own GI solution might never come out) if Unity decided to use/upgrade Enlighten even after 2024.

And Enlighten 4 improved a lot when combining with real-time ray tracing, so I assume that user devices should support ray tracing?

 

No build realtime GI that works out of the box without needing RTX GPU. And no, SSGI is not enough for that . Also would be good if worked in URP

 

I fully expect most if not all solutions to eventually require it once the hardware is ubiquitous. The Steam hardware survey is rapidly replacing many of the "G"s with "R"s. The baseline NVIDIA RTX card is now at fourth place.

 

RTX is still huge performance hit sadly, also there is a portion of market on AMD GPUs. Thats why I think the solution should support non-rtx stuff.

 

AMD GPUs support raytracing. Just not as fast as the NVIDIA cards.

 

Even if there will be realtime GI solution like RTXGI, I would like Enlighten 4 gets updated.

 

Frankly, if they backpedal even more and decide to go back all in with Enlighten I think I will lose the remainder of my hair.

 

In my opinion, Realtime GI makes your game better less than 10%
The biggest impact on the beauty of a game is the ambient light, the post processing, the AO and the real-time reflection.
Realtime GI solutions(like lumen) simulates the ambient lighting, AO and realtime reflections that makes the game beautiful. So the GI is not what your looking for. You must focus on a good ambient lighting, AO and reflection effects

Without GI in 2017:

 

I think it depends on what you want to do, if you want to focus more on realistic aesthetics, then you need GI, otherwise ambient lighting will do. But where it gets more difficult and I would like dynamic GI would be in dynamic scenes e.g. with voxel terrain. For example, the game Froundy uses a custom GI solution to make the caves really dark. The problem with their solution is that even if there is an overhang, it gets dark very quickly underneath.

In your examples, all are more or less stylized examples, and the example with the car doesn't seem realistic

 

If u are talking for outdoor environments than u are correct because there is very subtle effect caused by the bounced lighting in the outdoor areas and only major difference is made by a complete offscreen capable AO system which makes the forest region look believable, instead of washed out by the skylight, but for indoor lighting GI is necessary or else things would look very flat if your game wants photorealistic look

 

Interior:



Outdoor:

 

The indoor scenes are a bit flat here in terms of lighting, in the living room AO is also a bit strong (my impression) and the ambient light also illuminates places that should be a little darker (under the tables, behind the doors.) The person with the chair also looks kind of out of place, are they added to the picture later or are they part of the scene?

 

even though u have successfully created the illusion of photorealistic scene but if u compare this with a GI version or real image it will definitely look flat and off ... i would rather use these techniques on mobile devices... i highly doubt that the setting u used in the above images would be consistent across all outdoors and indoors scene (in an open world scenario)

 

The person with the chair is a scanned model. So has its own shadows by default

_________________________________

I just want to say that the GI is only less than 10% of the graphics. The post effects, AO and realtime reflection + a good textured scene is the 90% of the graphics

 

I agree that GI does not mean beautiful graphics, BUT

(dynamic) GI is essential for some use cases, such as emission.

Another example is ray-tracing. We can still have beautiful rendering without it, but some ideas will be difficult to accomplish.

 

Emission solution:

Tutorial

 

Is it Enlighten?

I think we all know Enlighten's limitations.

Anyway, Enlighten should not be here after 2024 and we are waiting for Unity's own GI solution.

Dynamic GI used to be a nice-to-have feature, but since it is existing in engines like ue and godot, I think we are expecting something more than Enlighten(3) and it should at least be important:

 

Not a bad effort. But I don't think it's going to scale well in a large game as there's a pretty constant hit in reading texture information. Works fast for one, but not for many.

But looking at this reminds me - we can use cookies to basically achieve the same result. I like your sprit of innovation, very inspiring. Keep up the good work. I also like that you are not afraid to try older techniques in a modern way.

 

It's actually the other way around. Ambient light is used to simulate/fake indirect light, and ambient occlusion is used to simulate/fake indirect shadows (Wikipedia refers to AO as a "very crude approximation to full global illumination"). Indirect lighting & shadows are handled naturally in global illumination because they are "consequences" of how light physically bounces around the scene.

The reason global illumination is important is because, just like shadows, it grounds objects in the scene to their surroundings, which is something that ambient light & occlusion can't do in every situation.

For your examples, you've conveniently chosen the exact type of scenes that benefit from ambient lighting & occlusion: scenes that are lit mainly by direct light and the colors of the major objects are similar. Try making a scene that's lit mainly by indirect light and the colors of the major objects are different.

This was just a test taken way back in Early Access 1 with some modified cvars.

One of the most unrealistic & stylized games, Minecraft, uses global illumination (light propagation) as a gameplay element.

 

The problem I see is that behind the chair it is just as bright as everywhere in the room. This reinforces the impression that it was added later using Photoshop.

 

Lighting is definitely not 10% of what can make a game look good though. Quake RTX, which has the same models and textures it had more than 25 years ago, looks better than most remasters of 10 years old games with entirely remade assets and textures but old lighting.

Unity needs a dynamic GI solution, and I hope APV will eventually lead to DDGI, but it probably won't be before years (if it ever does). Raytracing won't be a thing you can rely on (in any engine) until raytracing-capable hardware is used by vast majority of users, which won't happen for years either. Until then, we're stuck with Enlighten GI and all of its limitations...

 

DDGI won't happen, it's too memory intensive. Lerping APV sets is scalable.

 

What about APV with realtime raytracing probes?

 

Why do we still need lumen if most of the player's computers can handle RTGI in the feature?

 

I think lumen is supposed to be faster than RTGI in terms of performance

 

Lumen is still only workable in the same hardware tier that can run RTGI well. For example it's a bit of a pig on my 2070S. And that card does reasonably good RTGI.

Dies on 980 but the 980 can't do RTGI at all but it also really sucks at general GI.

The "without RT" part is not really the goal here. RT is not expensive. RT is very expensive. Which is true? well that depends on what the dev is asking it to do. For example, Unity's SSGI can have a very lightweight use of RT to fill in offscreen information and this would not change it's performance, or a heavy use of RT. Configurable.

https://docs.unity3d.com/Packages/c...ion@14.0/manual/Override-Screen-Space-GI.html

 

I haven't done testing myself tbh but i think you're right about the "without RT" part because my gtx 1050 basically cant handle lumen

 

I would like to clarify something about Realtime GI.

1. Wanting Realtime GI with anything below RTX2080 GPU @1080p, 60FPS is just madness. Doesn't matter Lumen, SSGI or anything, it is just not a viable hardware. It will work in a simple or very small scenes, but in a moderately size scene with moderate number of polys, eg 10mil, it just won't work. For example, a 1080 could probably handle minecraft or Counterstrike level of rendering with Realtime GI in GTX1080, but that will probably be the limit, even for Lumen. There are ways to get around it, but not while compromising either the quality or the performance of hardware accelerated RT.
2. The "without RT" part is regards to using hardware accelerated structure of a GPU, mostly nVidia. So it doesn't mean older cards can run it without a performance impact. It means equivalent cards of power from other vendors eg. AMD 6600XT for RTX2080 not GTX1080, GTX980. Nothing from this point on will give older card owners power, time goes on and newer cards will always be better. And rendering standards will move up. Remember GTX1080 was released in 2016. GTX980 in 2014. We have to be able to accept that.
3. Lumen has 2 modes, software lumen (A clever mix of VXGI, SSGI and who knows what) and hardware lumen (Raytraced). Unity also has 2 modes, SSGI and RTGI. Lumen's software version is far superior to that of Unity's SSGI, performance and quality wise. As for Raytraced, I doubt one will be better in terms of quality as they are both raytraced
4. The difference in my opinion, is that Lumen allows a combined, simple workflow when it comes to reflections and so far, Unity requires a lot of elbow grease. On top of this, Unity has been bad at rendering metallic things (Something is always off). But this is GI so I will move on.
5. Lumen, at the moment seems way better because Unity's SSGI is really crap. And DX12 on Unity which is required for RT, is really crap. I won't say Lumen is good because the opponent (Unity) failed to come to the fight. They just failed to come. So for now, I will say Lumen is the better choice or more like the only working choice. Unity's RT is actually fundamentally good, but as long as DX12 is broken like now, it won't matter.
6. We live in a strange time where RT makes every non RTX card the same and this is hard to swallow but the age of rasterization is slowly ending, at least, it won't be the golden standard anymore. Devs want to use raytracing for audio, spatialization, etc. We will end up using more raytracing in the future just like DSR won't go away.
7. A lot of devs say that GTX1060 is still the most popular GPU and hence that minimum requirements should have GTX1060 sitting comfortably in it, but we have to be honest. If you are running GTX1060 in 2022, then clearly you have given up on PC gaming. There are many factors to account for here: the rise in GPU prices, and availability. But even so, GTX1060 in 2022 is just to low for realtime GI. A sound business choice but that doesn't mean GTX1060 can handle Realtime GI, Lumen or RT regardless.
8. In case you are not aware, UE5 is not an upgrade to UE4. UE4 and UE5 is fundamentally different. UE4 is the superior version of current HDRP. UE5 on the other hand, is Epic's killer move to dominate the Open World Game Engine. They are getting rid of every baking aspect of the engine so that smaller studios can work on large open worlds. They are literally nurturing their future clients. It is frighteningly scary that they are this bold. Teens who are eager to try open world game dev will compare UE5 and HDRP and will choose UE5 without second thought. Because they want to try to make the next GTA. I mean they will end up throwing few boxes and capsule colliders but that is how they will enter the game dev world. It is clear that UE4 and UE5 will coexist for a very long time for these reasons. UE4 for traditional games, and UE5 for open world games.

I hope Unity starts a new RP, maybe OWRP (Open World Render Pipeline). I mean, this is half a joke, because I just don't expect Unity to take any of this seriously. but if I may keep on dreaming

• Float32 so that we are finally free from floating point issues
• Native streaming level even in editor
• AI Navigation that works on large scale
• GPU culling
• Ditch SRP batcher and make instancing standard.
• Terrain system from 2020.
• Fix DX12

I mean, what I listed above plus Nanite is pretty much UE5.

Lastly, I heard from one of the devs that the HDRP team is "much smaller" than we think and pretty much the team is short-handed 24/7. According to Google, there are about 5000 employees...

Five thousand and of that five thousand the HDRP team (the spearhead of Unity) is still very small and short handed...

Shouldn't there be at least 100+ team members dedicated to HDRP...out of 5000

Not counting document or support guys (no disrespect)

but 100+ solid devs to work on HDRP.

I thought this was a given...

I wonder how short handed the DOTS team is...

 

if we just leave terrain and srp batcher, isn't unity already improving all the others with DOTS and improving SRP batcher with BRG???



Coming back to the topic, I believe that unity has no plans to bring real-time GI solution anytime soon or they must have just abandoned as we haven't heard about it for a long time....if they are going to bring it out in next 2-3 years i believe RTX (or AMD's own real-time GI solution) would have become the standard solution.... If unity wants to bring Realtime GI, they must really target weaker hardware and even mobile phones (it would be okay if its going to be precomputed or similar to enlighten)

 

DOTS, from the looks of it will never make it in time. When I say time, I generally mean the working cycle of UE5, which I assume for the next 3 years. UE5's open world capability (all the listed) is already available, whereas DOTS....his parents just started dating. I am not a huge fan of DOTS and have received a bit of hate from DOTS lovers, but this is more of will it arrive on time rather than will it resolve the issue. In order for DOTS to be the solution to the listed features:

• DOTS needs to be production ready - > debatable, but probably will not be for the next year or so even if you are super optimistic. I personally see it 2+ years for it to reach feature parity of current Unity Gameobjects. DOTS 1.0 is slated for this year, but that won't mean everything else will be ready.
• DOTS hybrid renderer needs to be production ready -> clearly not happening soon. Even if so, subscene based approach is sort of outdated, but since it is pretty much on hold nowadays, I am just going to assume the best. Giving Unity the benefit of the doubt.
• Float32 is not DOTS related, and probably won't work. DOTS will suffer Floating Point Origin issues just like now.
• Everything else (Terrain, AI Nav, GPU culling, DX12) is not DOTS related and won't be effected. I just hope it doesn't break or not work with DOTS.

So no. DOTS is not really relevant to this topic. Hence why I see DOTS as something "not so important" but I'd rather wait for it than to bash it at this stage. Because Open World could still be a thing 3 years later and maybe not. I don't think the trend is going to die quickly though given the Metaverse thingy.

As for Unity's intention on Realtime GI. My guess is that they are eagerly waiting for RTX to become the standard. We can see it via the pathtracing and other raytracing features added since. It is actually not too bad. Except DX12 is really crap...and Unity devs seem like they don't want to believe that...go check out DX12 performance threads to see what is going on but basically, pretty much everyone is saying that DX12 performance is terrible, but the devs think it is "not that bad". Dunno, but I think it is horrible. Everyone expects DX12 to be faster to begin with

So unless DX12 meets DX11 performance soon (probably won't happen, as the devs are still convinced that DX12 is "ok") Unity's RTX features are moot. It doesn't exist. its like saying "I will give you 10 trillion dollars if you have 10 trillion dollars in your bank account". The prerequisite is just obnoxiously impossible.

 

when you guys talking about the performance of ray tracing, I will say my rtx 2060 can handle rtgi+rtao+rtr together with dlss and over 50fps. That is the performance when I playing cyberpunk 2077 with highest graphics

 

I'm pretty sure SSGI is not really limited by the amount of triangles/poly counts.

 

What you don't seem to understand or appreciate is... you can't throw more dumb people at a problem and still get a good answer. This isn't a lottery. The kind of people you need are extremely smart, rare and hard as hell to hire.

If it wasn't the case then Nasa would find new hires growing on trees. Quality of new hires is everything or you just get fatter, not taller.

 

Theoretically yes. But try in Unity and you will notice that it starts to degrade both in quality and performance.