What will come after Geforce 3?

[atavus]

Any ideas on what Nvidia will try to include on their next card? I know they probably haven't let any information out, but what are people speculating on?

<b>P</b>eople for the
<b>E</b>ating of
<b>T</b>asty
<b>A</b>nimals

 

[Guest]

From what I hear they'll probably improve the geomatry unit a bit for the GF3 Ultra, I'd also guess that they'll have to impliment N-Patches or ATI is going to have one hell of a geometry performance/resolution edge. After that I'd look into what Microsoft is trying to add to DirectX 9.

I AM Canadian.

 

[OzzieBloke]

Things we can expect in GeForce 4:

Twin Unified Realtime Data Streaming, or
TURDS

Power-User Sub-System Interface, or
PUSSI

Digital Inverse Control Kinematics, or
DICK

and

Multi-User Fully Functional Direct Interface Variable Enhancement Restructuring, or
MUFFDIVER.

Cow with legs spread wide either dead or playing 'cello.

 

[HolyGrenade]

Actually they won't include pn triangle (n-patches) support. They've already gone down the polynomial route in the GeForce 3. The GF3 has dissapointed a few people in its pushing power. It is more powerful than the GF2 series in most situations, and definitley will be in the DX8 games, but still it isn't that powerful. The GF3 Ultra should be much more powerful. It should be a performance upgrade.

ATI have also been saying the performance is already there in the current cards, the radeon 2 will be a feature upgrade. This leads me to think it won't be that much more powerful than the GF3.

The nVidia nv30 is next in line for the next major feature upgrade. It will be DX9 compatible although there have been rumours that they had a fall out with microsoft in terms of the features to be included in dx9.

DirectX9 should have features that weren't included in dx8 like 3D textures and some new features like NURBS amongst others.


<font color=red><i>Tomorrow I will live, the fool does say
today itself's too late; the wise lived yesterday

 

[noko]

Any idea when 3d textures will be used. The Radeon fully supports 3d textures and the sample programs from ATI are really cool. Talking about an ideal texture system, forget about LOD with procedural 3d textures, they look good close or far away with virtually no band width hit. N-Pathces are supported in DX-8 why whouldn't Nvidia support something already in the API? It makes the most sense in being supported and having a remarkable effect on object quality. Nurbs would be great, first true rounded objects but collision detection and basic physics calculations on board would be more important.

<P ID="edit"><FONT SIZE=-1><EM>Edited by noko on 07/05/01 11:38 AM.</EM></FONT></P>

 

[HolyGrenade]

I'm not sure I understand how LOD using 3D textures work. Do you have any further info or links?

Both, ATI and nVidia want higher order surfaces, and both will probably finally adopt to nurbs. But, right now nVidia wants to push everyone to start using higher order surfaces on hardware using the standard polynomial curve surfaces. ATI, on the other hand, is hoping most designers are already using polygonal surfaces compatible truform. So they want polygonal surfaces as input for the card.

Even though the current cards will both have different ways of taking in the 3D data, they will have to tesselate it in some stage. And frankly, I have no Idea how good these tesselation units are.

The problem with the truform method is that, for games coming out in near future (Unreal 2 & Doom 3, afterall these are the games the expensive hardware owners want to play), will have huge amounts of polygons. Doing truform on that will no doubt stress the hardware to max, and performance aint gonna be so good. It should, however, be ok with games prior to that, providing the 3D surface in the game is compatible.

The problem with the polynomial surface approach is that game developers need to make their models in curves rather than polygon models. The only Engine I know to be using curves is Quake 3 in its environment, but not the models. I'm not sure, there may be others. If the tesselation unit is good enough, and the system has enough memory, there is another probable solution. Any curves in the game can be tesselated by the CPU to low poly surfaces for collision detection and then the original curves be sent to the GPU for tesselation into high poly surfaces for display purposes. If, instead, the models are made of polygons and have a conservative polygon count, providing the CPU is powerful enough (about >1GHz), it should be possible to get curves from the polygons without too much of a performance hit. This should allow the original polygon model to be used for collision detection and the curves can be sent to the GPU for eyecandy, I suppose.

The best thing about the curves is that they can be used for scaling, i.e. dependant on your hardware and the required LOD. I do think onboard Collision Detection and some basic physics functions to determine the rebound and recoil level to simulate realistic collisions would be great. I've been ranting about that for ages. May be, we should send lots of emails to the graphics card manufacturers, microsoft and SGI and tell 'em what we gamers want!


<font color=red><i>Tomorrow I will live, the fool does say
today itself's too late; the wise lived yesterday

 

[noko]

<A HREF="http://www.ati.com/na/pages/resource_centre/dev_rel/sdk/RadeonSDK/Html/Samples/OpenGL/RadeonVolumeTexture.html" target="_new">3d texture using generated texture</A>

The texture is mathematically generated so LOD of a 2d texture is not used (no 2d texture is used). Meaning no matter how close or far away the object is the texture will look perfect. No texture mapping is needed therefore no mismatch overlaps will be seen on the oject. The 3d volume will have internal mapping of the texture which will be consistent with outside texture. Like real objects.

I agree collision detection and basic physic calculations in the GPU would be a new level for gaming and simulators where real interactions with the simulated enviroment could be done closer to virtual reality.