Unreal Engine 5, [UE5 Developer Availability 2022-04-05]

[Digidi]

Geometry doesn't determine how a surface looks when lighting bounces off it. That's determined by the multiple layers of texture surface information that can make it look like metal, or plastic, paint etc. Physically base rendering materials.

Normals are then used to add surface detail to create a sense of depth to the texture when appropriate, like rust on metal being slightly extruded, which when combined with proper lighting further enhances the illusion of reality.

Fine details in geometry certainly combine with the lighting to create a far more detailed and real scene, but it's the combination of this, PBR materials and a modern lighting solution which sells the illusion. You seem to be focused only on geometry as the means to achieve these results.

Geometry

None of those small geometric details would actually cast shadows discernible were not for UE5 new VSM scheme and SS-Shadows. Better geo gives lighting better cues, but better lighting makes that better geo actually pop. Neither loooks good enough in isolation.

There is more like shadows. You have light reflections which highly depends on the angel of the surface which the lights is hitting. Or the light which is pasing a ruff edge on acorner will look different like a streight edge.

I did test differnt lighning. I don't see any big different between lumen and screen space her. But what i see are details which the geometry bring on this game.

 

[Shifty Geezer]

If you look at the train station, it is not true that there is no complex gemoetry and that is my point. It is the realy small geometry and the Shadows of this, wich make the difference. The shadows don't have to be 100% precice or raytraced but they must be there.
Look at the small gaps of the uneaven bars. This make the seans authentic. If the walls are streight even with raytracing it will look like plastic.

And yet real. Did you miss my post, or just ignore it?

We can reduce the discussion to a couple of simple examples that show your argument doesn't hold. The first is the Cornell Box. It doesn't matter how much micro-geometry you add to the flat planes to introduces irregularities - it'll only look realistic with a realistic lighting model. And for any complex geometry you might care to pick, imagine it running on GS Cube with PS2 level lighting... it'll never look authentic. This railway station won't look real thanks to incredible details if rendered with a few points lights and hard shadows.

Your complaint is more that RT as used isn't making the difference, not that good lighting doesn't make a difference. True realism needs realistic lighting, realistic models, realistic materials, and realistic animations. A weakness in any of these comes across, but it's lighting that makes a scene look real like a physical thing in the real world or not. The other areas affect whether the scene looks legitimate or fake.

 

[dobwal]

I think he's saying that is there's no need for the fancy ray traced lighting (Or Lumen) on next gen as without the increase in geometry detail it makes little to no difference.

Tell that to minecraft.

 

[PSman1700]

Tell that to minecraft.

Or Metro… or chernobylite, 2077 or…. most games with RT lighting.

 

[Digidi]

Tell that to minecraft.

Minecraft is a bad example it didn't have any proper light system in the original version.

Compare Cyberpunk or other RT games. Everybody says that the Visual improvement are not very much. And than look past games and UE5 Games. Even UE5 gaves have no Raytracing but Lumen and achive simliare light quality because the geometry catch it. Every Graphic Jumps was always high Poly jumps. Like Crysis or UE5.

 

[PSman1700]

UE5 does employ RT though.

 

[Kaotik]

UE5 does employ RT though.

Yes, it does. But his point stands, the difference isn't exactly mind blowing between SW and HW Lumen

 

[dobwal]

Minecraft is a bad example it didn't have any proper light system in the original version.

Compare Cyberpunk or other RT games. Everybody says that the Visual improvement are not very much. And than look past games and UE5 Games. Even UE5 gaves have no Raytracing but Lumen and achive simliare light quality because the geometry catch it. Every Graphic Jumps was always high Poly jumps. Like Crysis or UE5.

Yeah, probably because in a world thats totally composable by the player, fake lighting is hard to implement. And games that have great non RT lighting probably used RT during development to create reference scenes which were used to light their games.

Lumen (non RT hardware) is a form of software based ray tracing and isn't cheap. It requires 8ms at 1080p on consoles.

Lumen Technical Details in Unreal Engine | Unreal Engine Documentation

Lumen provides two methods of ray tracing the scene: Software Ray Tracing and Hardware Ray Tracing.

• Software Ray Tracing uses Mesh Distance Fields to operate on the widest range of hardware and platforms but is limited in the types of geometry, materials, and workflows it can effectively use.
  
  
• Hardware Ray Tracing supports a larger range of geometry types for high quality by tracing against triangles and to evaluate lighting at the ray hit instead of the lower quality Surface Cache. It requires supported video cards and systems to operate.

 

[pjbliverpool]

Yes, it does. But his point stands, the difference isn't exactly mind blowing between SW and HW Lumen

SW Lumen is a form of RT.

 

[Kaotik]

SW Lumen is a form of RT.

True, should have specified he accelerated RT.

 

[function]

Every Graphic Jumps was always high Poly jumps.

I don't think this is remotely true.

Texturing had an enormous impact on 3D graphics, allowing for perceived detail that was far, far beyond anything geometry alone could achieve. Lighting (via any and all methods) produced atmosphere and a sense of objects existing in a 3D world. Per pixel shading of materials had a transformative effect on graphics, making even low poly models present an enormous amount of meaningful information to the observer.

Skilful use of muti-texturing (e.g. base, detail, specular, normal) along with good shaders allowed results better (more successful at getting the representation across) than you could have achieved simply by using crippling amounts of geometry.

Graphics are produced via a multitude of techniques and properties, and to pick geometry out and put that ahead of everything else is very misguided IMO.

 

[Digidi]

Textures are ok, but why do you use textures? Normaly you don't need texturs when you have poloygon of the size of an pixel than each polygon has only a colour. Textures where envented because they couldn't render everything with small pologons. With textures you will never achive the depth information which you can achive with polygons used for complex structures.

 

[function]

Textures are ok, but why do you use textures? Normaly you don't need texturs when you have poloygon of the size of an pixel than each polygon has only a colour. Textures where envented because they couldn't render everything with small pologons. With textures you will never achive the depth information which you can achive with polygons used for complex structures.

No. This is a fundamental misunderstanding you have, and one that demonstrates your ignorance on the subject of graphics.

You could have polygons the size of a pixel, and they wouldn't through their triangular boundaries and normals be able to show things like back scattering, SSC, varying specular across materials etc.

At best, what you're proposing would be horrendously inefficient and also shit.

 

[Jawed]

I don't think Reyes was ever considered shit

Reyes rendering - Wikipedia

 

[OlegSH]

Textures are ok, but why do you use textures?

Because rather than store color and attributes per vertex, it's likely more efficient to store them as separate attlases (textures) with UV parametrization from compression point of view.
This way, you would likely achieve much higher compression rates in case of multiple textures per object with the same parametrization. It's also much more flexible in how you can combine textures / shaders / meshes vs keeping them baked into geometry.
Rather than giving up on textures, there is actually more sense in doubling down on them, because even geometry benefits greatly from storing local low precision displacement values in textures (with great compression) rather than full fat world space coordinates without compression, this way you can quantize geometry and store it in way more compressed form via tesselation with displacement mapping.
Also, games always had lower geometry density vs texture density, still remains to be truth even with the Nanite, so keeping parametrization and great compression of textures is still very relevant even with Nanite.

 

[Dampf]

There's a new compiled Matrix demo using a newer engine version (5.1)

Apparently, it is using mesh shaders for Nanite. In my testing it does run noticeably smoother.

Most interestingly are added high quality transparency reflections for glass, as seen on cars. Window reflections look so much better than before and there is no performance impact from enabling them (Yes, I tested that without the frame cap at 99% GPU usage). Awesome stuff!

 

[Shifty Geezer]

Textures are ok, but why do you use textures? Normaly you don't need texturs when you have poloygon of the size of an pixel than each polygon has only a colour. Textures where envented because they couldn't render everything with small pologons. With textures you will never achive the depth information which you can achive with polygons used for complex structures.

You're arguments are non-sequitur (you don't need atomic level detail to get photo realistic rendering; texture aren't in any way a restrictive approximation; the example you are using of this photorealistic station demo is using textures and not geometry at the level you are talking about needed to replace texturing), and furthermore you aren't engaging in the debate in balanced fashion but only using the closed mindset of having a theory, presenting it, and then presenting arguments you think support your theory without following the arguments that challenge your theory.

You haven't considered my example of the very best geometry you can think of rendered at PS2 level lighting. Are you honestly saying that'd be realistic graphics? Why isn't the original Toy Story "photorealistic" in accurately portraying simple plastic models then, where the level of detail is fine for the subject matter? You haven't addressed the issue of the Cornell Box, the reference which can contrast a real world reality against computer graphics where geometry is unimportant.

Mod hat on: At this point we're just generating noise. If you aren't going to change the style of the discussion to something more open, I'll end this 'agree to disagree' talk to stop this thread getting unduly polluted and to save people wasting time in dead-end discussion.

 

[iroboto]

Minecraft is a bad example it didn't have any proper light system in the original version.

Compare Cyberpunk or other RT games. Everybody says that the Visual improvement are not very much. And than look past games and UE5 Games. Even UE5 gaves have no Raytracing but Lumen and achive simliare light quality because the geometry catch it. Every Graphic Jumps was always high Poly jumps. Like Crysis or UE5.

The jump is in the lighting. As lighting continues to improve it will get closer to real life lighting. As for; it’s not doing much more, I would disagree entirely. It’s costly to make massive steps to improve lighting so the steps are smaller, but they are still progressing and now we have progressed into the beginning of RT dynamic lighting marking the transition point of leaving baked lighting behind.

Improving geometry and deciding to be stuck on baked lighting, versus the other way around; it wouldn’t be long for people to see the error of that decision.

 

[London Geezer]

There's a new compiled Matrix demo using a newer engine version (5.1)

Apparently, it is using mesh shaders for Nanite. In my testing it does run noticeably smoother.

Most interestingly are added high quality transparency reflections for glass, as seen on cars. Window reflections look so much better than before and there is no performance impact from enabling them (Yes, I tested that without the frame cap at 99% GPU usage). Awesome stuff!

Those windows! It was my biggest gripe with the first demo. Great stuff.

 

[Andrew Lauritzen]

I did test differnt lighning. I don't see any big different between lumen and screen space her. But what i see are details which the geometry bring on this game.

Did you try turning off Virtual Shadow Maps? The differences there are pretty stark IMO.

CSM:


VSM:


The CSM case still benefits somewhat from the additional geometry, but it's harder to tell the difference between texture/normal detail and real geometry without it affecting the fine-grained shadows as well.

Another example. CSM:


VSM:


In the case here it's very obvious how the loss of detail in shadows affects the perception of depth significantly.