While that is correct, the method that Nvidia and ATI use are not compatible with a majority of algorithm's that are widely used on the market.
The tessellator is fully programmable with a hull shader before tessellation (to determine the amount of tessellation and do control point setup etc), and a domain shader after tessellation (to process/adjust/correct the triangles generated by the tessellator).
You can implement a large variety of algorithms this way, including very popular algorithms such as bezier/b-spline and Catmull-Clark.
Perhaps not EVERY possible algorithm can be implemented at this point, but that certainly doesn't mean the tessellator isn't very useful.
Tessellation does not improve "Image Quality". Image quality can only be improved by a better texture. It is useless to have tessellation enabled with a very low texture as it really shows the image in bad view.
Nonsense.
Firstly, you're assuming that textures are used in the first place. That is not necessarily true (although I agree it is rare in games).
Tessellation adds geometry detail, which WILL improve IQ, as surfaces can be made smoother/more detailed.
Secondly, texture quality is not really the breaking point in current games. They already contain VERY high-detail texture maps, bumpmaps etc. These are generally wrapped around relatively lowpoly objects, and shading (bumpmap/parallax) is used to create the impression of geometric detail.
Tessellation can replace this by REAL geometry, which also will be MSAA'ed properly, unlike bumpmapping approaches. Again, IQ improvements.
That it like having a perfectly created 3d human with a really bad texture. In reality, a better texture is better than 3d geometry as it is easier on the video card. It just needs more/faster memory, that is why 2gb or even 4gb will be here in the future vs fully utilized DX11 tessellation techniques.
In case you don't realize, a very common approach for tessellation is to use displacement maps (textures) to 'encode' or 'compress' the geometry.
It's not mutually exclusive. Thing is however, that displacement maps can be more compact than bumpmaps/heightmaps/horizonmaps that we need today, while at the same time delivering better image quality (see above: better MSAA etc).
Tessellation is not faster as a whole. If you have so much geometry that it cripples a GPU then what is the point of implementation?
That's if you look at it from the wrong side.
The point is not in "generating as much geometry as possible", but in rendering a certain amount of geometry as efficiently as possible. Given a fixed amount of geometry, tessellation will always be faster, as polycounts increase, because you reduce the memory footprint and bandwidth requirements.
In fact, the main problem with tessellation you are taking something that can be done on the CPU and are now moving to the GPU. Go talk to Blizzard or any other MMO creator and you will see that people have better CPU's. When the hardware from top of the line to the bottom has enough dedicated hardware to use tessellation, only then can I recommend it.
Tessellation can be done on the CPU... but there are two problems:
1) The GPU has far faster dedicated tessellation hardware. Even the lowest-end DX11 part versus the highest-end CPU (just like even the fastest CPU can't outperform even the simplest Intel IGP in software rendering in general).
2) Tessellation is done in realtime (as it is adaptive: tessellation factors depend on distance and/or screen space size of the polygons). This means that for every frame, you need to re-tessellate every object. The net effect of that is basically that your CPU needs to do all the geometry processing, generate tons of data, and then try to push it over the PCI-e bus to the GPU. We've had hardware T&L since the first GeForce, and this is exactly why: you cannot push the geometry around fast enough.
Really, trying to argue for CPU-based geometry processing was naive back in the early GeForce days... but today it's just ridiculous.
I think you shouldn't have made that post. I get the distinct impression that you just want to argue against tessellation because of brand loyalty. You throw some technical terms around, but have no idea about how it works in practice.
Facebook
Twitter