REYES vs. Shader Model 4.0

[j^aws]

What are your guestimates on Shader Model 4.0 and how will this unified shading language differ/ share with REYES and it's single shading stage Thanks in advance...

 

[nAo]

REYES hasn't a single shading stage

 

[MfA]

Im 99% certain SM4 will still be based on vertices and fragments ... which isnt a huge deal if the vertex shading is flexible enough to generate new vertices on its own (ie. it can do adaptive subdivision) and the hardware is unified. You could do as much in the vertex shader as you wanted. I doubt SM4 will be that flexible though.

With Reyes you loose too much control over subdivision IMO.

 

[no_way]

Im 99% certain SM4 will still be based on vertices and fragments ... which isnt a huge deal if the vertex shading is flexible enough to generate new vertices on its own

Wouldnt it be then actually based on primitives and fragments, i.e. you cant do subdivision on vertices but you can on primitives.
If so, i hope it will have other primitives apart from triangles ( and tristrips and trifans )

 

[MfA]

Well that is just statefull vertex shading

 

[Sage]

i want voxels, dammit!

 

[GeLeTo]

REYES is a rendering alghorithm - like raytracing, scanline rendering and z-buffer rendering.
Maybe you mean the Renderman Shading Language?
And what's that with the single shading stage? I'd say renderman has more stages - dicing, displacement, lighting, shading...

 

[j^aws]

GeLeTo and nAo,

Sorry guys, I thought I was clear...by single stage I mean shading is only done on vertices of micro-polygons rather than vertex and fragment stages.

REYES is a rendering alghorithm - like raytracing, scanline rendering and z-buffer rendering.
Maybe you mean the Renderman Shading Language?

Yeah kind of...what I'm trying to imply is that changes to SM1, SM2, SM3 and SM4 are approaching the offline renderer, REYES, with each iteration, no?

 

[Remi]

Rendering micro-polygons is not strictly speaking dependant on the shading language level but on the capability to rebind a buffer on the card (using a pixel shader to output geometry packed in a render target buffer, and reusing that buffer in a second pass as a vertex buffer).

While that can do wonders, it certainly consumes bandwidth (a lot!), which is why I don't think we'll see that soon in real-time graphics. Using detail maps (parallax mapping for instance) is a lot more efficient (bandwidth wise) to add visual details.

Just my 2 cents however...