GDC2004:DirectXNExt Graphics Future

[Nv40]

FOund this infor very interesting.. that came from GDC2004..
by mr.charmansteve


2004-2005 (DirectX 9.0c / Shader Model 3.0)
Small Batches
Geometry Instancing
Vertex Texture Read
Floating Point Blending
Pixel Complexity
Dynamic Flow
Centroid Sampling

2006+ (DirectX 10 / Windows Longhorn)
New Driver Model
State Refactoring
Virtualization
Remove Fixed Function
More Resources
Render to Vertex Buffer

2007+
NURBS
Raytracing
Subdivision Surfaces
Radiosity

http://www.watch.impress.co.jp/game/docs/20040326/dx1003.jpg#
http://www.watch.impress.co.jp/game/docs/20040326/dx1004.jpg#
http://www.watch.impress.co.jp/game/docs/20040326/dx10.htm

exciting eh? :drooling:

[OldOfEvil]

NURBS!


wasn't NUON supposed to be all about the NURBS?

[Viral]

Isn't PSP (playstation portable) supposed to use NURBS? and that's set for 2004.. i had no idea the tecnology was so far from PC.

[MUYA]

question...what is NURBS? :p

[Geforce4ti4200]

whatever happened to tile based rendering? didnt the kyro cards have that? from what I heard, tile based rendering is very cpu intensive so youd need the best cpu for the card to shine

[zoomy942]

i had a kyro ii and the idea of tile based rendering ios what made me buy it. it was a great idea.

[Morrow]

question...what is NURBS? :p

Non Uniform Rational B-Splines are used to create smooth/round/natural 3d models in most all professional 3d programs because the base of this technique are not polygones but curves. NURBS most likely used to model 3d characters, especially faces but not limited to this.

Since the technique is not build on polygones current GPUs cannot accelerate NURBS.

Further developpement has resulted in NURMS (Non Uniform Rational Mesh Smooth) with is basically the same as NURBS but with no UV quadratic restriction with means texturing the 3d models is much easier.

[nutball]

2007+
Raytracing
Radiosity


Note the question marks against there -- I'm pretty sceptical we'll see these in a hurry.

[Remi]

Since the technique is not build on polygones current GPUs cannot accelerate NURBS.

Further developpement has resulted in NURMS (Non Uniform Rational Mesh Smooth) with is basically the same as NURBS but with no UV quadratic restriction with means texturing the 3d models is much easier.
Just to add my 2 cents, it looks like the idea behind those slides are just to use the topology processor to tesselate the NURBs (create an optimized triangle meshes in real-time from the NURBs). This relies on the capability of a shader to create triangles, something we can't do today - however there's a trick... I haven't tried it myself, but I think one could do NURBS today in a vertex shader by feeding the shader with flat triangles meshes that it'll modify, instead of creating them ex-nihilo. You only need then to be sure you're sending enough triangles, and the shader can kill the unused ones...

One problem remains however before using this in games: programming efficient collision detections with nurbs isn't really as easy as it is with triangles...

[Morrow]

Just to add my 2 cents, it looks like the idea behind those slides are just to use the topology processor to tesselate the NURBs (create an optimized triangle meshes in real-time from the NURBs). This relies on the capability of a shader to create triangles, something we can't do today - however there's a trick... I haven't tried it myself, but I think one could do NURBS today in a vertex shader by feeding the shader with flat triangles meshes that it'll modify, instead of creating them ex-nihilo. You only need then to be sure you're sending enough triangles, and the shader can kill the unused ones...

There are indeed some tricks to accelerate curves like NURBS and splines since every curve can be rendered using unlimited straight lines (polygones) but the workload is quite high.

One shouldn't forget that Carmack already managed to hardware accelerate splines in Quake3 (which he called "curved surfaces") with almost perfect collision detection so NURBS aren't really that far off but there still some technical limitations left which need to be taken care of.

EDIT: Although NURBS are great for professional modeling I would rather see hardware accelerated radiosity in games than NURBS support :)

[Remi]

Thanks for the info, I was under the impression that only some restricted subset of nurbs were allowed in Quake2.
Although NURBS are great for professional modeling I would rather see hardware accelerated radiosity in games then NURBS support :)I agree 200%. :p

EDIT: I meant Quake3, of course. (That was 3, right?)

[Nv40]

Thursday 25th December 2003
CVGames:
new advances in DirectX. Where do you see such technology taking us, in terms changes/advancements it will allow developers bring to the FPS genre?

Brian Jacobson, senior software engineer at Valve:
The continued movement toward increased general programmability of the GPUs is going to result in more developers making more varied and interesting special effects ever more quickly. A bunch of algorithms that currently must be done on the CPU, such as facial animation, will be able to be accelerated by hardware soon. We're going to see continued evolution and improvement of shadowing solutions, and far superior lighting solutions like high-dynamic range rendering (HDR). HDR is already possible, but we'll see better implementations made possible by new hardware available within a year. Screen space effects, allowing us to do things like motion blur, depth-of-field, colour correction (as done in movie film processing) and tone mapping, will become commonplace. Some developers are also interested in taking advantage of algorithms made possible by dramatic increases in fillrate we see coming down the pipe, such as deferred rendering, which allows us to do things like decals on normalmaps.

http://www.computerandvideogames.com/r/?page=http://www.computerandvideogames.com/news/news_story.php(que)id=99674






John carmack:Id software.. Feb 8 2003,
The future is in floating point framebuffers. One of the most noticeable thing this will get you without fundamental algorithm changes is the ability to use a correct display gamma ramp without destroying the dark color precision.Unfortunately, using a floating point framebuffer on the current generation of cards is pretty difficult, because no blending operations are supported, and the primary thing we need to do is add light contributions together in the framebuffer. The workaround is to copy the part of the framebuffer you are going to reference to a texture, and have your fragment program explicitly add that texture, instead of having the separate blend unit do it. This is intrusive enough that I probably won't hack up the current codebase, instead playing around on a forked version.
http://www.webdog.org/plans/1/



i highlighted some of the interesting things according with valve that will be possible in NExtGeneration hardware.. im curious about the -more advanced HDR rendering- that VAlve is waiting to see in next generation hardware. i have read that SM3.0 with Floating Point Blending greatly enhances HDR. is the NV40 already there in specs with the hardware that Valve was waiting? also it looks that the lack of FP blending is a major limitation that JC face in the first generation of DX9 cards from ATI/Nvidia for using correct display of gamma ramp.. at least here its very likely that Nvdia released the hardware that will make life easier for Idsoftware. :)

hardware acceleration of facial animation and DEpth of field seems very interesting too. wonder if those things will be possible or not in the Nv4x/R4xx hardware.

[Drew Carey]

EDIT: Although NURBS are great for professional modeling I would rather see hardware accelerated radiosity in games than NURBS support :)
I'm gonna have to agree. It's pointless to allow NURBS support for ingame enviornments at the moment.
Seems to me that they're adding this inorder to phase out the cpus from the proffesional rendering process and start working entirely with graphics card.

[bloodbob]

Opengl has had nurbs support for ages the only problem is on the windows platform the NURB code is M$ code and is rather pathetic.

Anyway what do you think is easier nurb support or raytracing??? I think the answer is damn clear.

[AngelGraves13]

ok....what does all this mean for us gamers? DirectX Next is gonna be like a leap from DirectX 5 to 9?

[Remi]

Not really. In fact... I think the most important part of DXNext is... well... actually... I'm not sure...

Because, as I see it, there's lots of excellent things in DXNext, but they're more on the technical side and they have an effect on the gamer, but an indirect one.

For instance, it'll bring a kind of multitasking and virtual ressources to the GPUs, but that doesn't really gives a clear and definite advantage over what we have now. Sure, you'll be able to open several games at the same time on your PC and to switch from one to another. Sure, a cheap low-end graphic card will run the latest games with all options maxed but it'll be at slide-show speed. Is it a good thing? Sure! Is it a clear and definite advantage? Hum...

Sure, VS40 and PS40 means the chips will be "more active", or in other words they will adapt themselves better with their work load... Is it a good thing? Sure! Is it a clear and definite advantage? Hum...


With programmability, the most notable things will come more and more from the software side and less from the hardware one. Well, that's not to say the hardware won't matter - it will a lot! - but just that the innovation won't be in the hardware alone but also increasingly in the way it's used: software.

So the question is... how will it be used? What will developpers cook with it?

Beware, highly speculative ground ahead... :cool:

[Nv40]

ok....what does all this mean for us gamers? DirectX Next is gonna be like a leap from DirectX 5 to 9?

Not.exactly.
The diferences between Dx9 to Dx10 are more Improvements ,Tweaks ,better ways to implement things ,few IQ improvements and *FAR* greater PRogramability.

However its also true ,that far greater programability ,easier ways to enhance the IQ in games ,also helps to improve the graphics in games,since Developers will be able to get the graphics they want more quickly , and greater programabily allows more complexity in easier ways without killing the performance of your video card. in other words greaterprogramability+fasterperformance=better graphics.

So thats why every new ShaderModel version or Directx version always helps developers in one way or another to add more realism in their games (albeit sometimes in an Indirectly way),since it makes their lives easier. and developers creativity are not only limited by their talent but also ,by performance and by a tight schedule ,they need to deliver the game in Time.

according to this table


2004-2005 (DirectX 9.0c / Shader Model 3.0)
Small Batches
Geometry Instancing
Vertex Texture Read
Floating Point Blending
Pixel Complexity
Dynamic Flow
Centroid Sampling

2006+ (DirectX 10 / Windows Longhorn)
New Driver Model
State Refactoring
Virtualization
Remove Fixed Function
More Resources
Render to Vertex Buffer

2007+
NURBS ?
Raytracing ?
Subdivision Surfaces ?
Radiosity ?



it will not be until 2007? that we will see the next big jump in graphics ,which is the main diference that separates Professional Graphics ,(holywood like) with games. Real time raytracing/radiosity (not hacks done in pixel shaders) are the next big thing that will allow natural lifelike Lights in games. If they are implemented in 2007? in realtime ,then it will take 4-5 years for the first few games to show it,but first you will see it in small ATi/NV demos. this fits well with Johncarmack prediction of Lord of the Rings graphics in games in ~10 years. :)

Subdivisions/Nurbs are not Image quality features at all ,they will not be noticed by the gamer.they are better tools for gamedevelopers ,that makes *far easier* for game developers the creation of games when it comes to Modeling/texturing .