ATI Ultra-Realistic Realtime Ruby Tessellation demo

[Kakkoii]

Video:
http://www.youtube.com/watch?v=y4uUx051zBk


Textured:
http://www.flickr.com/photos/a...ed/3907249086/sizes/o/

Untextured:
http://www.flickr.com/photos/a...ed/3907250684/sizes/o/

Normals:
http://www.flickr.com/photos/a...ed/3906472189/sizes/o/

Specular Highlights:
http://www.flickr.com/photos/a...ed/3907249780/sizes/o/



Quite amazing. I hope they make this as a released tech demo for the 5000 series.

 

[IntelUser2000]

Why do I have a feeling that DX10 maxed out can do this...

 

[dguy6789]

I don't think it's a matter of just being able to produce the end result image, but being able to produce that image much much faster and more efficiently than before.

 

[Kakkoii]

Originally posted by: IntelUser2000
Why do I have a feeling that DX10 maxed out can do this...

DX10 can't completely do this. As you need support for general computing on shader's to run the ray tracing code. DX11 has the ComputeShader function for doing general computing on a GPU's shaders.

Also DX10 only supported 4k x 4k textures, which this model is probably using much higher than. DX11 supports up to 16k x 16k. Plus there's the better multi-threading support for CPU's in DX11. And the new texture compression technique.

There's a bunch of factors in DX11 that helps something like this be rendered in real time, with ray-tracing on top of it none the less.

But the ray-tracing probably isn't run through DX11. It's most likely run using the same type of coding that goes into something like Folding@Home utilizing the stream processors on your GPU.


Oh, and almost forgot the most important part.. Tessellation. Notice how the image of the mesh is almost completely solid? That's because of Tessellation. Normally that highly detailed of a mesh would lag your computer to fuck. Never mind freaking ray-tracing it and applying an incredibly high-res texture. (Yes ATI could do Tessellation before, but now it's in DX11, made better, and where market share of ATI/Nvidia isn't deciding it's fate.)

 

[LittleNemoNES]

eww... way too pasty

 

[MrK6]

She is amazing; best tech demo I've seen yet. This brings up an interesting point - as the capabilities of hardware get move and more advanced, are programmers actually going to take the time to use them? The hours spent for that tech demo alone most be insane.

 

[AdamK47]

Originally posted by: dguy6789
I don't think it's a matter of just being able to produce the end result image, but being able to produce that image much much faster and more efficiently than before.

Wasn't that the claim for DX10?

 

[jimhsu]

I'm optimistic about tools like ZBrush cutting on development times by an order of a magnitude, despite the realism of models like these. Dynamic noise generation, fractal geometry, etc all was designed to tackle making ultra high resolution models for games and art. And now we can actually render these in real time.

 

[ShadowOfMyself]

The texured pic is crazy... Finally a model that looks realistic

I think most people wouldnt be able to tell thats not a real person if she was photoshopped into a background

Now thinking about the previous Ruby versions that looked so... hum.. plastic, what a difference

 

[Idontcare]

Originally posted by: gersson
eww... way too pasty

You know I was thinking this too, sorta...it's like Ruby has aged right alongside AMD all these years.

She's beautiful for my tastes, reminds me a lot of the refined distinctions my wife has acquired in the past 10 yrs as well, but not exactly the renderings of an early twenty-somethings vixen either.

(I'm specifically looking at the face-shot in this high-rez image where they added skin blemishes to improve the photo-realism)

I'm thinking this is telling us something about the demographics data that AMD marketing is seeing when it comes to PC gamers...our average age must be heading towards mid-thirties and fast or we'd be looking at Ruby's much younger sister being rendered up on that screen.

 

[LTG]

Wrong.

Not only could DX10 do this, DX9 could also. It's doing it really fast, but that's mostly because the card has a lot of raw horsepower.


Here is an example of real time ray tracing using Adobe's version of a pixel shader on top of opengl, which is really not different from a pixel shader on DX9:
http://mike.newgrounds.com/news/post/156863

Tess. hardware and DX11 compute can be used, but are not required at all.

That said if features are easier to use developers are more likely to use them.

Originally posted by: Kakkoii

Originally posted by: IntelUser2000
Why do I have a feeling that DX10 maxed out can do this...

DX10 can't completely do this. As you need support for general computing on shader's to run the ray tracing code. DX11 has the ComputeShader function for doing general computing on a GPU's shaders.

Also DX10 only supported 4k x 4k textures, which this model is probably using much higher than. DX11 supports up to 16k x 16k. Plus there's the better multi-threading support for CPU's in DX11. And the new texture compression technique.

There's a bunch of factors in DX11 that helps something like this be rendered in real time, with ray-tracing on top of it none the less.

But the ray-tracing probably isn't run through DX11. It's most likely run using the same type of coding that goes into something like Folding@Home utilizing the stream processors on your GPU.


Oh, and almost forgot the most important part.. Tessellation. Notice how the image of the mesh is almost completely solid? That's because of Tessellation. Normally that highly detailed of a mesh would lag your computer to fuck. Never mind freaking ray-tracing it and applying an incredibly high-res texture. (Yes ATI could do Tessellation before, but now it's in DX11, made better, and where market share of ATI/Nvidia isn't deciding it's fate.)

 

[soccerballtux]

Originally posted by: Idontcare

Originally posted by: gersson
eww... way too pasty

You know I was thinking this too, sorta...it's like Ruby has aged right alongside AMD all these years.

She's beautiful for my tastes, reminds me a lot of the refined distinctions my wife has acquired in the past 10 yrs as well, but not exactly the renderings of an early twenty-somethings vixen either.

(I'm specifically looking at the face-shot in this high-rez image where they added skin blemishes to improve the photo-realism)

I'm thinking this is telling us something about the demographics data that AMD marketing is seeing when it comes to PC gamers...our average age must be heading towards mid-thirties and fast or we'd be looking at Ruby's much younger sister being rendered up on that screen.

That's an interesting theory.

That image brings the uncanney valley to mind.

 

[gorobei]

hate to break it to you, but that's a hi-rez laser scan of a real person.

the normal map is a static directional instead of a tangent map, that means it's functionally a statue with no deformers beyond maybe vertex displacement.

the demo is meant to show that they can realtime light and dolly around a (i'm guessing billion poly object)model. there may even be some raytracing in there along with the subsurface scattering for the skin.

 

[jimhsu]

Suffice it to say that I think most modern engines (e.g. Crysis) are somewhere in uncanny valley territory right now. Most of the scenes are amazingly realistic at ultra high detail settings, yet you tend to focus on the differences from reality and get distressed because of those differences. Unfortunately game AIs are nowhere near that territory right now -- that still needs serious work.

 

[dflynchimp]

Originally posted by: jimhsu
Suffice it to say that I think most modern engines (e.g. Crysis) are somewhere in uncanny valley territory right now. Most of the scenes are amazingly realistic at ultra high detail settings, yet you tend to focus on the differences from reality and get distressed because of those differences. Unfortunately game AIs are nowhere near that territory right now -- that still needs serious work.

the sad truth is we are still quite a few years from seeing "true" AI. Current AI models are still based on scripted reactions that don't stand up to close scrutiny.

The DX11 renders looks pretty damn good. I wonder if the demo has any potential as a game engine or at least potential as a development tool for next gen game engines.

 

[Damn Dirty Ape]

kinda hot. the demo technology too. heh

 

[MODEL3]

beautiful eyes

 

[Philippart]

wow, incredible!!! and I thought the new generation wouldn't show any special advantage over the last apart from more performance...

 

[AmdInside]

Meh. Not impressed. Show me this in a video game or real world demo, and I'll be impressed. I am excited for DX11 much more than I ever was for DX10 though.

 

[IntelUser2000]

I'd be sad if that kind of graphics reaches games and first thing I see is non-realistic movement and characters getting stuck and jittering.

 

[Kakkoii]

Originally posted by: gorobei
hate to break it to you, but that's a hi-rez laser scan of a real person.

the normal map is a static directional instead of a tangent map, that means it's functionally a statue with no deformers beyond maybe vertex displacement.

the demo is meant to show that they can realtime light and dolly around a (i'm guessing billion poly object)model. there may even be some raytracing in there along with the subsurface scattering for the skin.

Yes it is a high-rez scan using the Light Stage technology that's used in many big movies. But they are probably using tessellation a long with the extra hardware to make it run so fast. Or else this could have been done merely with a 4870x2.

Here's a discussion about the technology AMD did:
http://www.youtube.com/watch?v=wBbWH_m1Re4

 

[OVerLoRDI]

IMO it looks too realistic

 

[Modelworks]

Originally posted by: jimhsu
I'm optimistic about tools like ZBrush cutting on development times by an order of a magnitude, despite the realism of models like these. Dynamic noise generation, fractal geometry, etc all was designed to tackle making ultra high resolution models for games and art. And now we can actually render these in real time.

The only way you will render top level Zbrush models in real time in games for now is if that model is the only thing on the screen. The biggest benefit of tessellation is that I can make one model for a game at a modest poly count . The people with high end hardware can use tessellation to make it look better. No more levels with pipes made out of 8 sides shapes or heads that are 16 sided primitives.

 

[nemesismk2]

The Tessellation demo looks very nice, if they can use it with games then that will be great.

 

[jimhsu]

Originally posted by: Modelworks

Originally posted by: jimhsu
I'm optimistic about tools like ZBrush cutting on development times by an order of a magnitude, despite the realism of models like these. Dynamic noise generation, fractal geometry, etc all was designed to tackle making ultra high resolution models for games and art. And now we can actually render these in real time.

The only way you will render top level Zbrush models in real time in games for now is if that model is the only thing on the screen. The biggest benefit of tessellation is that I can make one model for a game at a modest poly count . The people with high end hardware can use tessellation to make it look better. No more levels with pipes made out of 8 sides shapes or heads that are 16 sided primitives.

Right, that's why tessellation, normal maps, etc that kind of stuff was invented. So that people with low-end systems can see simplified models, and people with really high-end system can see models that resemble (but are not) the original.