AMD demos DX11 hardware at QuakeCon 2009

[Heinz68]

The card itself looks very short. This can't be there top of the range model surely. Plus it looks as if it only has one PCI-e connector.
Provided it's not Photoshop, It must be the mid or entry level card. Little funny is that CHIPHELL (http://translate.google.com.au/translate?u=http%3A%2F%2Fwww.chiphell.com%2F2009%2 F0820%2F99.html&sl=zh-CN&tl=en&hl=en&ie=UTF-8) copyrighted the pictures but have no idea what model it is.

[jlippo]

Actually if you do use compute shaders for rendering you can render what ever you want from spheres to more complex surfaces 'perfectly'.
If you can do it in math you can do it in compute shaders.

[Yaboze]

Speaking of Quake and ATI/DX11......

Isn't Wolfenstein OpenGL though?

[pakotlar]

Speaking of Quake and ATI/DX11......

Isn't Wolfenstein OpenGL though?

nope its D3D9 for SP, and OpenGL for what I hear is awful MP

[pakotlar]

What he means is, all "curves" in the GPU are approximations, either using triangles or straight lines(in the case of 2d shapes), there is no true curves, though honestly, I dont see the requirement for them, since upto a certain point, you cant tell the difference between a true curve and an approximation

Yeah this is absolutely correct. There isn't a real need for it beyond efficiency gains. If someone figured out how to build a rendering engine that could simply output NURBS or other spline-based curves, and on the animation and modeling side someone figured out how to do good UV texteruing and bones or other fancy control-point animation or dynamic animation systems for spline models, efficiency gains would be HUGE. If the graphic card was able to just output NURBS or splines without tesselating them of course. Anyone who has worked with 3d modeling programs can attest to how much more processor intensive it is to render a finely tesselated mesh compared to the original nurbs model. It's orders of magnitude slower (I'd say at least 10x at the point where it is relatively indistinguishable). The truth is that polygons are only really around because they are easy to work with, and are reasonably fast when they are large and there isn't a lot of overlap, or you have great occlusion culling techniques. Beyond that its a **** show.

Triangles are not the way of the future; they are just the best adaptable tech we have today. It's like LCD's or plasma tech. Neither is particularly good to start with, but by investing a lot of resources into them you can catch up in the areas they suck at. Editable poly's are like that. They have great flexibility (the thinness of the LCD) but they also have great processing cost (the slow response rate, poor color accuracy).

[pakotlar]

What he means is, all "curves" in the GPU are approximations, either using triangles or straight lines(in the case of 2d shapes), there is no true curves, though honestly, I dont see the requirement for them, since upto a certain point, you cant tell the difference between a true curve and an approximation

The other problem is the "indistinguishable" point requires the use of high-frequency triangles ergo micro-poly's which have atrocious efficiency rates. This is part of the reason why Xbox 360 has a ceiling of ~ 500 Million triangles, but doesn't ever render more than 90-140 (140 is generous). The bandwidth, and computing costs of setting up half a billion tiny as f.u.c.k triangles, then doing occlusion, AA, whatever else to them is lunacy.
aside: I know that Xbox 360 can technically render something like 2.5 times that. The setup engine is limited to 500M (and for a good reason, as it would be idiotic to waste transistors doing something that in reality cannot be used except in corner cases).

Next generation you'll see maybe ~ 300M, which would account for 5M/frame at 60fps, which IMO is an incredible amount. Crysis, on very high (with some tweaks) on a GTS 285 gives us about 100M @ 1680*1050 or 1440*900 [depending on the tweaks] (I'm being very generous, assuming 2.5M/frame at about 45fps). Although it seems that there are some new techniques for dealing with the efficiency of micro-poly issue as seen at Siggraph 2009.