What is more important next gen, Vertex or Pixel processing?

[j^aws]

Just wanted to gauge peoples views on whether increases in vertex (polygon) processing or pixel processing will be the key for better 3D quality games next gen

 

[AlNom]

both equally: relatively high polygon characters + fancy shading = UE3.0

 

[David_South#1]

Voted: Pixel Processing.

As technology appears to me, this is how I view it. i.e. IMO

Once technology it at the point of producing more polygons per frame [at a rate exceeding 72hz] than even the highest relevant display format has pixels, pixel quality become more important.

I understand that there are different types of polygons with multiple sizes and shapes. That FSAA and super sampling must also be considered in making such a comment. So the number has to be higher than this general view. But I also understand that computer graphics cards are now theoretically producing over a billion polygons/triangles a second. So it is my opinion the first consideration has been met.

At this point developing better lighting models/algorithms and improving our understanding of how the eyes interpret things is the overriding concern. It will likely be managed by how well a pixel looks. So texels and other effects affecting pixel surfaces are IMO more important.

 

[Qroach]

I voted both, as vertex shading and pixel processing will be known simply as "shader" in the future. Since they are merging, it will be hard to say one is more important than another IMO.

 

[ERP]

Obviously only my opinion, but while instruction counts in vertex shaders will rise, it will only be a marginal change. Pixel shaders will have 10's, possibly 100's of times the instruction count that they had in the XBox era.

 

[Fafalada]

Vertex of course, once you have enough of them, you don't need any pixel operations at all except "paint single color"

 

[AlNom]

Vertex of course, once you have enough of them, you don't need any pixel operations at all except "paint single color"

It's official, you're insane.

 

[ERP]

Vertex of course, once you have enough of them, you don't need any pixel operations at all except "paint single color"

And your going to solve the geometry aliasing problem how?

 

[MfA]

Obviously only my opinion, but while instruction counts in vertex shaders will rise, it will only be a marginal change. Pixel shaders will have 10's, possibly 100's of times the instruction count that they had in the XBox era.

On the other hand, the average pixel shader's complexity is depressed by the amount of pixels used to determine shadowing. As for geometry aliasing, you guys will have to find out how to use dynamic geometry LOD ... no matter how much of a bitch it is. Your worry, not mine

David, you cant light a pixel unless you know if it is actually being lit ... and you cant determine that in a pixel shader alone.

 

[Fafalada]

And your going to solve the geometry aliasing problem how?

That kinda depends on how you'd generate the geometry in the first place - I wasn't exactly thinking of just having it all stored explicitly.

 

[ERP]

And your going to solve the geometry aliasing problem how?

That kinda depends on how you'd generate the geometry in the first place - I wasn't exactly thinking of just having it all stored explicitly.

That's not really the problem, once polygons get into the <2 or 3 pixel range you get some hideous aliasing when things are in motion along color discontinuities give it a try. Supersampling can improve the issue, but you need a lot of samples to match even average texturemapping.

Pixar solve the problem by tessalating in screen space which is what texture mapping effectively does.

To be honest I see your point, I just don't see the point of throwing away a perfectly good solution (texturemaps) and trying to solve the problem a different way for the sake of it.

 

[j^aws]

And your going to solve the geometry aliasing problem how?

......
......
To be honest I see your point, I just don't see the point of throwing away a perfectly good solution (texturemaps) and trying to solve the problem a different way for the sake of it.

Faf is a Rebel! But shouldn't programmers always try something differrent if there's a chance of solvinng the solution with the same computing costs, or at least for the sake of experimentation which could lead to other avenues which may never have ever been explored?

 

[archie4oz]

I voted vertex for fun, but I'm more along the lines of "both"... Even with screen space tessellation of meshes, texture maps will unlikely ever go away if not for anything than to store functions, arbitrary lookups, displacements, etc...

 

[Squeak]

Obviously in the future, pixelshading is going to take up a lot more die space than it does now. Next gen. no one is going to be satisfied with hardwired gouraud only. The future pixelshaders are going to be so powerful and flexible, that maybe hardwired shading can be ditched, making todays attempts at P.S. look like useless toys.
That said, a pixelshader needs vertex data to work on, else it just a powerful 2D computer.
So both.

 

[Tahir2]

We are already on to Phong Shading in the PC space I believe.
For next gen of consoles importance will lie with Pixel Shading on XBOX2 and with Vertex Shading on PS3. Gamecube2 is a complete mystery to me.

At least that is what my crystal ball is telling me

 

[Squeak]

We are already on to Phong Shading in the PC space I believe.

I know it's theoretically possible to do phong on todays hardware, but are any games really doing it? I suspect almost all xbox games are using regular linear (gouraud) interpolation of the normalmaps.
Am I wrong?

 

[MfA]

Gouraud uses linear interpolation inside the tri of the shading results at the vertices themselves ... if it uses per pixel shading it isnt gouraud anymore period.

 

[Squeak]

Gouraud uses linear interpolation inside the tri of the shading results at the vertices themselves ... if it uses per pixel shading it isnt gouraud anymore period.

I didn't mean that it was gouraud, only that tangent space normal mapping interpolates the normals across the polygon, in the same linear fashion gouraud interpolates colours.

 

[nAo]

As for geometry aliasing, you guys will have to find out how to use dynamic geometry LOD ... no matter how much of a bitch it is. Your worry, not mine

That's the same thing I repeat all the time..trying to sound like a broken record
We need to switch to 'different' geometry representations..ok, that's the last time I write that on B3D..I promise!

 

[London Geezer]

As for geometry aliasing, you guys will have to find out how to use dynamic geometry LOD ... no matter how much of a bitch it is. Your worry, not mine

That's the same thing I repeat all the time..trying to sound like a broken record
We need to switch to 'different' geometry representations..ok, that's the last time I write that on B3D..I promise!

I suggest you just put it in your signature...