• Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.

Killzone 4 Uses Raytracing For Its Real-Time Reflections

Status
Not open for further replies.
P

phosphor112

Banned
shaderhacker said:
One last point I'd like to bring up.

The only way to prove for sure that these are real area lights is to see the lights move in realtime. Most of these games are using static lights on static objects. They could be pre-baked.

Also, it would be very easy to mimic area lights with creating several point lights all in an array. Like this:

****
****
****
****

I just made a rectangular area light with 16 point lights. :D

In the conventional sense, this is not a true area light...

-M
Click to expand...


I don't think you understand what you're talking about regarding the demonstration that GG did.
kPyAKk8.png

yBFmlMv.png

7B12Utq.png

oWCYCUq.png

dByQTBK.png

KXTPOzS.png
Click to expand...
Clearly they are area lights.
 
P

phosphor112

Banned
shaderhacker said:
The BRDF on the surface has to generate it's distribution of rays (seen in the video presentation) based on it's model (diffuse has one set of samples that are completely hemispherical and the specular will be based on the lighting model - Blinn, Phong, etc..). Also the BRDF has to be evaluated at the light source as well. This is called importance sampling.

http://en.wikipedia.org/wiki/Importance_sampling

-M
Click to expand...

The noise still exists because it doesn't have enough ray samples. Importance sampling is just an algorithm to find out which rays it should be casting instead of just sending out random rays.

Random casting on an image that has the same casts as a render with importance sampling will always be worse. (i.e. random cast render < importance sampling render).

You can still increase the cast amounts to increase the quality even more, thus reducing noise.
 
S

shaderhacker

Banned
phosphor112 said:
I don't think you understand what you're talking about regarding the demonstration that GG did.

Clearly they are area lights.
Click to expand...

All those materials are highly reflective surfaces. Very easy to index a reflection map texture and blur it.

GG never told explicitly how they do it... because it's not the conventional way of ray-casting and sampling the surface. It's just a trick.

-M
 
P

phosphor112

Banned
shaderhacker said:
All those materials are highly reflective surfaces. Very easy to index a reflection map texture and blur it.

GG never told explicitly how they do it... because it's not the conventional way of ray-casting and sampling the surface. It's just a trick.

-M
Click to expand...
You're wrong.

It's real time. What is your problem dude. Earlier in the presentation they said how they had to prebake reflections into everything for KZ2 and 3, and for SF they don't have to do that don't do that because of the engine itself doing these things in real time.
 
S

shaderhacker

Banned
phosphor112 said:
The noise still exists because it doesn't have enough ray samples. Importance sampling is just an algorithm to find out which rays it should be casting instead of just sending out random rays.
Click to expand...

I know this. But it's a perfect indication of doing it the right way - trace rays.
 
S

shaderhacker

Banned
phosphor112 said:
You're wrong.

It's real time. What is your problem dude. Earlier in the presentation they said how they had to prebake reflections into everything for KZ2 and 3, and for SF they
don't have to do that
don't do that because of the engine itself doing these things in real time.
Click to expand...

I didn't say it was pre-baked. I said that it was using a texture lookup. Have you ever heard of a deferred shading renderer? GG uses that.

http://en.wikipedia.org/wiki/Deferred_lighting

Every single pass they do is considered "baked" into texture maps to be fed to the pixel shaders to compute lighting in 2D space. Not real 3D space.

My problem is that you guys think KZ4 can magically do real area lights when it requires ray-tracing to be done the conventional way.

-M
 
G

Graphics Horse

Member
shaderhacker said:
I didn't say it was pre-baked. I said that it was using a texture lookup. Have you ever heard of a deferred shading renderer? GG uses that.

http://en.wikipedia.org/wiki/Deferred_lighting

Every single pass they do is considered "baked" into texture maps to be fed to the pixel shaders to compute lighting in 2D space. Not real 3D space.

My problem is that you guys think KZ4 can magically do real area lights when it requires ray-tracing to be done the conventional way.

-M
Click to expand...


Not counting gbuffer lookups (why bring that up?) It's not using a texture look up except for the textured stained glass light obviously.
They explain what they're doing in the video for the specular reflections even though it's too complicated for me, the basic idea is simple enough.
http://www.youtube.com/watch?v=_29M8F-sRsU&t=54m27s
But they don't explain how that would interact with objects moving partially in front of the light, not very well probably.
 
P

phosphor112

Banned
shaderhacker said:
I didn't say it was pre-baked. I said that it was using a texture lookup. Have you ever heard of a deferred shading renderer? GG uses that.

http://en.wikipedia.org/wiki/Deferred_lighting

Every single pass they do is considered "baked" into texture maps to be fed to the pixel shaders to compute lighting in 2D space. Not real 3D space.

My problem is that you guys think KZ4 can magically do real area lights when it requires ray-tracing to be done the conventional way.

-M
Click to expand...

I'm not sure what you're getting at.

Are you saying that the buffers are 2D that these are somehow aren't real time area lights?

The lighting and glossy reflections are done in real time using the SSRT pass using the BRDF. It is later blended with the local cubes and skybox.

The lights are calculated in real time in 3D and then gets "baked" into a 2D screen space. The gloss reflection mask is 2D space if I understand what he's saying.

Maybe I'm just not understanding you? But...? I don't even know what I'm thinking anymore.
 
K

KKRT00

Member
shaderhacker said:
That exact light was one that I specifically looked for in C3. Couldn't find it.
Click to expand...
Its a showcase of tech, it wouldnt make sense in game is similar form, but the lights, that the guy from community day event were showing, are ingame. Every Area Light in CE 3.5 have the same properties.

shaderhacker said:
LOL! Basically you are saying if a game developer somehow made a light 'simulate' an area light, then it's an area light. OK. Let's see how many of these *dynamic* area lights make it into games this year.

-M
Click to expand...
Do You even know how Crytek's engine works? They do not prebake stuff or fake stuff, if they would want to fake this, they would just fake it with point light and wouldnt call it area light and make such differentiation in tech talks.

We really need CE 3.5 SDK asap.

shaderhacker said:
One last point I'd like to bring up.

The only way to prove for sure that these are real area lights is to see the lights move in realtime. Most of these games are using static lights on static objects. They could be pre-baked.

Also, it would be very easy to mimic area lights with creating several point lights all in an array. Like this:

****
****
****
****

I just made a rectangular area light with 16 point lights. :D

In the conventional sense, this is not a true area light...

-M
Click to expand...
And again to bring this up. You clearly havent watch those presentation, because they were talking about this. The point of making Area Lights is to avoid using many point light to fake it. Why? Because its cheaper to have one Area Light than 2-30 point lights.


====
sp3000 said:
That light is not in the actual game.
Click to expand...

No, it isnt, but Crytek doesnt fake this stuff, if they used Area Light shown in tech demo, You're certain that other Area lights behave in exactly the same way. Its philosophy of their engine, to have everything real-time and as much modifiable.
 
M

MarkV

Member
shaderhacker said:
I didn't say it was pre-baked. I said that it was using a texture lookup. Have you ever heard of a deferred shading renderer? GG uses that.

http://en.wikipedia.org/wiki/Deferred_lighting

Every single pass they do is considered "baked" into texture maps to be fed to the pixel shaders to compute lighting in 2D space. Not real 3D space.

My problem is that you guys think KZ4 can magically do real area lights when it requires ray-tracing to be done the conventional way.

-M
Click to expand...
Ah, now i know what you are talking about. I've probably bad news for you :) ...almost everything in real time rendering is about cheating.
So if Crytek and Guerrilla claim to be able to have dynamic light source that came not from a point light but from something that can have a rectangle-like shaped area that they want and with corrected specular highlight then they have all the rights to call this an area light ...even if it's not like area light from VRay for example, guess why :)

Sorry for my english.
 
P

phosphor112

Banned
I don't see why he'd think it's not real time when the guy already said Area lights are 15% more intensive than point lights. Why would it be more intensive it was just a point light with a texture draw call for the reflection?
 
S

shaderhacker

Banned
phosphor112 said:
I don't see why he'd think it's not real time when the guy already said Area lights are 15% more intensive than point lights. Why would it be more intensive it was just a point light with a texture draw call for the reflection?
Click to expand...

Because to make an area with points, you'd have to have several point lights to make up one area light.

I'm not saying it's not real-time. I'm saying it's not being done in the most accurate way possible - which is by ray-tracing and sampling the surface. Something that videocards are not good at doing. They are good at computing a 2D grid of threads using the same math operator. In order to get the GPU to be fast, you have to put the data format into something it can churn out quickly. Putting data into a 2D grid where the threads can each take several rows of pixels is ideal for GPU. But you can only get but so much accuracy with a 2D approach.

Ray-tracing will require a full 3D dataset pattern.
 
S

shaderhacker

Banned
I just looked at the latter part of the presentation.

That's pretty neat that they are able to get as far as they have. They are limited to spherical and planar area lights. It's also still not clear how they do shadows projected from area lights.
 
G

Graphics Horse

Member
shaderhacker said:
Because to make an area with points, you'd have to have several point lights to make up one area light.

I'm not saying it's not real-time. I'm saying it's not being done in the most accurate way possible - which is by ray-tracing and sampling the surface.
Click to expand...

I don't think anyone argues with this.
His repeated use of the term 'raytracing' is obviously going to rub people up the wrong way when none of it refers to 'real' raytracing.
 
M

Metroid-Squadron

Banned
shaderhacker said:
That exact light was one that I specifically looked for in C3. Couldn't find it.



LOL! Basically you are saying if a game developer somehow made a light 'simulate' an area light, then it's an area light. OK. Let's see how many of these *dynamic* area lights make it into games this year.

-M
Click to expand...
Remember that graphics techniques in games are only approximations that look like the intended effect. So sure, the area lights in KZSF aren't true area lights, but for their intended purpose they work just fine.

Also, can anybody get some video of area lights in the retail version of Crysis 3? Would be nice to compare.

-----------

The thing that bothers me with these implementations is that they're work for spotlights and not omnilights. But I guess it's better than nothing.
 
D

Dictator93

Member
Metroid-Squadron said:
Remember that graphics techniques in games are only approximations that look like the intended effect. So sure, the area lights in KZSF aren't true area lights, but for their intended purpose they work just fine.

Also, can anybody get some video of area lights in the retail version of Crysis 3? Would be nice to compare.

-----------

The thing that bothers me with these implementations is that they're work for spotlights and not omnilights. But I guess it's better than nothing.
Click to expand...

I will spend some time later tonight or tomorrow morning and get some shots of the area lights. A really good example is in that warehouse section. The light is coming generally from the ceiling... and not through tons of point lights.

Ill find others too in the indoor sections.

Also the semantics about being real or fake is just silly. Until we are running some sort of path tracing or something... everything is "faked."

I must agree though that the GG usage of the word ray tracing is very disingenuous.
 
S

squidyj

Member
phosphor112 said:
Do you mind using imgur as the image host? Abload is blocked at my work. It's hard to comment on things when I can't see them lol



Yeah, I'm not really understanding his argument either. The noise he mentions is because of the raytracing sample not being high enough. Not because of the BRDF, well that's my understanding when I messed with pre-rendered scenes.
Click to expand...

absolutely, noise occurs because path-tracing and all it's variants, as well as photon mapping are monte carlo algorithms which don't give the 'correct' solution in any single pass but instead eventually converge to an unbiased solution to the rendering equation over many many samples.

Which is also one reason why their performance is shit and they are inappropriate for games or other real-time applications at this time. And considering their scaling, unless we get quantum GPUs, possibly ever.


I will repeat that I want to know what level of error was introduced with their linear decomp :l. TELL ME DAMMIT.
 
A

AznSniper

Member
If you guys want further confirmation that Guerrilla is working on a new IP, here is a guy's LinkedIn that points him to working on both Killzone: Shadow Fall and the new IP.

x4eUBBH.jpg
 
JCreasy

JCreasy

Member
Was anyone else disappointed in the forest level at E3?

I wanted to see more urban environments with more cool lighting effects. That forest level sapped my enthusiasm.
 
T

Tabular

Banned
KKRT00 said:
Yes, reflections in KZ:SF are screenspace like in every other game.
Click to expand...

Yes, screen space. No, not like all other games. They are using a full screen raytracing pass to calculate reflections - not 'just' cube maps. They said so specifically. Screen space in this case is a geometry limitation; where geometry behind you is not being calculated and therefore not visible by the rays. They're technique appears to be a bit beyond the Crysis 3 version which also used raytracing in a limited way. Seems to look sharper here as well, but we'll have to see the full game before judging the limitations.
 
D

Dictator93

Member
Tabular said:
Yes, screen space. No, not like all other games. They are using a full screen raytracing pass to calculate reflections - not 'just' cube maps. They said so specifically. Screen space in this case is a geometry limitation; where geometry behind you is not being calculated and therefore not visible by the rays. They're technique appears to be a bit beyond the Crysis 3 version which also used raytracing in a limited way. Seems to look sharper here as well, but we'll have to see the full game before judging the limitations.
Click to expand...

Actually this sends out less rays than the C3 implementation. Both are screen space techniques that combine with cubemaps.

One thing that is better in the KZ version is that particles and motionblur appear to be added into the reflection due to this effect being done last in the chain of effects. I like that quite a bit.
 
K

KKRT00

Member
Dictator93 said:
Actually this sends out less rays than the C3 implementation. Both are screen space techniques that combine with cubemaps.

One thing that is better in the KZ version is that particles and motionblur appear to be added into the reflection due to this effect being done last in the chain of effects. I like that quite a bit.
Click to expand...

Yeah, C3 omits particles, but in C2 DX11 upgrade particles were shown in reflections
http://www.youtube.com/watch?feature=player_detailpage&v=ON8fU2RWyK4#t=70

--
Hmmm tested it. C3 does reflect particles with SSR, but not on water, like water had different reflection algorithm even with SSR on.
C3 does not reflect alpha textures, at least i couldnt spot it on any surface

Here is gif showing particles reflecting on 4 surfaces http://i.minus.com/iUhehpMD61aR3.gif
 
RoboFu

RoboFu

One of the green rats
There are a few different means to ray tracing in gaming. One is just tracing a single vector to get an accurate depth and/angle of something. Another is sending a vector out at an angle to see what it hits.
 
KaiserBecks

KaiserBecks

Member
Bojji said:
I don't get why people listened to this bullshit 11 years ago. Sony fanboys will believe anything.

Killzone uses normal SSR.
Click to expand...
But with 8GB ddr5 everything was possible back then! Infinite power! You‘ve got to hand it to them, they were always good at making people believe this stuff. Same with the ssd this gen. Still waiting for a game to look as good as that desert tech demo.
 
Status
Not open for further replies.

Similar threads

Sony could soon automatically play portions of games for you (the feature is often already used in many MMORPG games)
Platform 2
83
3K
-Minsc- replied
Resident Evil 4 Remake uses 4 DRMs on PC, at the same time
Analysis 2
94
10K
Mr Moose replied
The industry is really missing that magic that used to come with E3 week and live press conferences.
2
75
2K
cireza replied
Does FORZA HORIZON 5 get "on track" RAYTRACING? Or is this a falsely added label?
15
2K
Skifi28 replied
Most Anticipated Announcements of 24’
2
81
4K
DAHGAMING replied
Top Bottom