Dont bash curved surfases just becouse patches doesnt do what you want them to.
I don't think it's bashing it. Curved surfaces have their problems too, like they don't add detail. Curviness isn't detail, it's smoothness, and even though it looks a lot better with low-poly models most of the time, subdivision doesn't add much to a model over 8000 polys. Very often you see yourself adding polygons that are not smooth, but tend to have detail (hard edges) in them.
John Carmack pointed out the texturing problems, but there are loads more, like level of detail, texture coordinates and material seams. Triangles are about as stupid primitive that you can have as a graphics primitive, but they are extremely convenient. For example, their bounding volume is bound by their control points, they are discrete (i.e. can be detached and manipulated individually) and they can be textured in any way possible.
I have the feeling that most graphics primitives (such as curved surfaces, maybe not including subdivision surfaces) are going to lose from triangles.
Put it simply: complexity of geometry tended to be a problem in 3D graphics, but with the extreme bandwidth and pipeline increases we see today, the stream of triangles is becoming less of a problem. A good LOD algorithm can get the triangles/pixel ratio down to around 1 which will be a very feasable ration in a very short while.
Most of the added processing time nowadays goes towards shadowing, detail maps, projective lights, motion blurring/oversampling/other frambuffer tricks, but it seems the triangle limit is diminishing.
All of these new techniques get easier with triangles. With adaptive subdivision these techniques get worse to implement. Try to do animated soft-skinned subdivision surfaces with adaptive subdivision and shadow-casting.
Triangles are stupid, but in the end, all graphics cards draw triangles. I think that curved surfaces won't be all what they were hyped up to be, they simply add too little at too much cost.
Slashdot Mirror
I don't think it's bashing it. Curved surfaces have their problems too, like they don't add detail. Curviness isn't detail, it's smoothness, and even though it looks a lot better with low-poly models most of the time, subdivision doesn't add much to a model over 8000 polys. Very often you see yourself adding polygons that are not smooth, but tend to have detail (hard edges) in them.
John Carmack pointed out the texturing problems, but there are loads more, like level of detail, texture coordinates and material seams. Triangles are about as stupid primitive that you can have as a graphics primitive, but they are extremely convenient. For example, their bounding volume is bound by their control points, they are discrete (i.e. can be detached and manipulated individually) and they can be textured in any way possible.
I have the feeling that most graphics primitives (such as curved surfaces, maybe not including subdivision surfaces) are going to lose from triangles.
Put it simply: complexity of geometry tended to be a problem in 3D graphics, but with the extreme bandwidth and pipeline increases we see today, the stream of triangles is becoming less of a problem. A good LOD algorithm can get the triangles/pixel ratio down to around 1 which will be a very feasable ration in a very short while.
Most of the added processing time nowadays goes towards shadowing, detail maps, projective lights, motion blurring/oversampling/other frambuffer tricks, but it seems the triangle limit is diminishing.
All of these new techniques get easier with triangles. With adaptive subdivision these techniques get worse to implement. Try to do animated soft-skinned subdivision surfaces with adaptive subdivision and shadow-casting.
Triangles are stupid, but in the end, all graphics cards draw triangles. I think that curved surfaces won't be all what they were hyped up to be, they simply add too little at too much cost.
Michiel