Pixel or Vertex more important, looking forward?

[hovz]

Like a set of lowres and highres textures, a game could also include a set of lowres and highres meshes, catering for various levels of 3d hardware.

This does have problems with normal mapping, specifically that tangeant-space normal mapping (required for most compression schemes, and often best for animated geometry) is invalid if the model is changed.

But still, games have had LOD for geometry for quite a long time. As an example, the Unreal engine has supported geometry LOD since around the days of the original UT. City of Heroes is another game that supports geometry LOD. It's just that it's typically used mostly to scale back the geometry of objects that are far away, not so much to reduce polycount on low-end hardware (however, it would pretty much do that automatically anyway, since lower resolution would naturally mean lower geometry LOD).

the geometry lod in ut is a joke. it removes a tree or a pillar here and there. havent played city of heros, but the screen shoots look terrible.

 

[hovz]

Well, I don't want to destroy the nice flamewar here... but I think both sides are saying pretty much the same, but approaching it from a different angle.

Nope we aren't and I'll elaborate...

I believe one side stresses on the quality of textures and AA/AF to increase realism... And the other side stresses on geometric detail.

Wrong again. I want both and not just one part increasing, preferably at the same time and gradually. Texture filtering and/or Antialiasing alone will not bring any real progress as you won't come close to realism with higher geometry only. What on earth happened to happy mediums?

But it seems that the others interpret it as trading texture/image quality for geometry.

Partially yes. I just don't want the concentration to fall on just one department. I would want a lot of things too, yet I'm also at least attempting to understand natural restrictions and indirect measurements that sadly still exist in the graphics market in order to come to more balanced conclusions.

However, if you look at the polycount today, and say, 4 years ago, it has pretty much stagnated, and that's rather sad. Agreed, the same polycount looks better today, because of the better textures and normalmaps and pixelshading, but still, the silhouettes are blocky, and animation is a tad limited. So yes, I would be happy if this area were explored, and texture quality and resolution remained at the current level for now.

The real first big game that contained a pure T&L optimised code was UT2k3. How many years after the introduction of T&L units on GPUs? It always takes time. I personally was quite disappointed in the past from almost non-existant dx8.1 shaders support a couple of years back, while on the other hand hope started to rise again with the appearance of the R300 and developers quickly adopting even some dx9.0 class shaders.

It might sound foolish from me to put too much hope on WGF, yet it's real target IMHO is to really leave only AI and physics to the CPU, with the rest falling onto GPUs. On chip adaptive tesselation isn't just a gimmick feature, dynamic LOD, geometry compression, stencil shadows calculated w/o the CPU and the list could go on and on. I don't think early WGF hardware will bring a gigantic revolution either, yet it at least sounds at this point of time a step into the right direction.

Another rather simplistic sounding question: why are developers lately instead of picking real displacement mapping via VS3.0 rather opting for parallax mapping? (1) It's cheaper to implement and probably delivering higher performance too and, (2) it can be realised on a much wider variety of accelerators out there.

As for high resolutions, it's currently only a side measure for me to decrease partly in combination with other techniques general aliasing or other annoying patterns. If we ever would get as far to render games that would look damn close to Pixar's Finding Nemo as an example, resolution would be the least of my concern.

your whole train of thought is flawed. you support developers coding for the lowest hardware, yet you expect them to suddenly change when wgf is introduced? that will be even more advnaced then what we have now that is being wasted. 2005 will be no different than 2004. graphics will remain stagnent, as nothing in the forseeable future offers any step foward.

 

[hovz]

again, when sony releases ps2 games and ps1 users have to upgrade to play the newest games, do they alienate their users? for 300, the price of a new console, u can buy a new video card.

i still laugh when i hear developers and publishers complain about bootlegging. they expect us to shell out money for their shitty cookie cutter games that are nothing more than a rehash of a console port or previous game.

 

[KimB]

halo2 looks better than any of the current unreal games in every aspect.

Except that you can play UT2004 at higher resolution, with better FSAA. Plus it appears to have higher polycounts and higher-resolution textures. It doesn't have the normal mapping, but that's because it's based on older engine technology.

 

[KimB]

i was unaware everyone buys a new pc when a new windows comes out.

Once again, idiocy from you.

Dumbass, the vast majority of people don't ever upgrade the operating system they get with the PC they purchase. So most people won't ever have a problem with the increased system requirements of Windows because it will have come with the PC.

 

[hovz]

i was unaware everyone buys a new pc when a new windows comes out.

Once again, idiocy from you.

Dumbass, the vast majority of people don't ever upgrade the operating system they get with the PC they purchase. So most people won't ever have a problem with the increased system requirements of Windows because it will have come with the PC.

they also dont complain when they cant utilize all the new features of newer windows. which is everyones arguement about developers no tmaking more advanced games.

btw u rly shouldnt mention idiocy. have you forgotten your entire history of posts during the introduction of the geforce fx? had the forums not been erased a few months back id have some fun digging up your old posts. you are quite possibly one of the dumbest individuals ive ever seen

 

[hovz]

halo2 looks better than any of the current unreal games in every aspect.

Except that you can play UT2004 at higher resolution, with better FSAA. Plus it appears to have higher polycounts and higher-resolution textures. It doesn't have the normal mapping, but that's because it's based on older engine technology.

and halo2 still looks better, it might techincally have highe rpolygone counts, altho thats debatable, but it certainly doesnt look like it. you just proved my point tho, all pc gaming offers as of late is higher res with aa af. if that doesnt seem like a waste of 3 to 500 bucks to you, then enjoy pissing away your money

 

[KimB]

the geometry lod in ut is a joke. it removes a tree or a pillar here and there.

No, it doesn't. It reduces the geometry of actors dynamically depending upon how far away they are. There is no removal of objects from rendering, beyond those that are out of sight. What the hell have you been smoking?

 

[KimB]

they also dont complain when they cant utilize all the new features of newer windows. which is everyones arguement about developers no tmaking more advanced games.

Using new Windows features wouldn't help developers make more advanced games.

btw u rly shouldnt mention idiocy. have you forgotten your entire history of posts during the introduction of the geforce fx? had the forums not been erased a few months back id have some fun digging up your old posts. you are quite possibly one of the dumbest individuals ive ever seen

Heh, go for it. The forums weren't erased.

 

[hovz]

the geometry lod in ut is a joke. it removes a tree or a pillar here and there.

No, it doesn't. It reduces the geometry of actors dynamically depending upon how far away they are. There is no removal of objects from rendering, beyond those that are out of sight. What the hell have you been smoking?

i was referring to dynamic lod in conjunction with world detail. but the dynamic lod functions in exactly the same way. for example if u put the lod to low it just wont display the wings of the manta once you reach dustance a. it doesnt dynamicly alter the complexity of the models, but im sure u already knew that

 

[hovz]

they also dont complain when they cant utilize all the new features of newer windows. which is everyones arguement about developers no tmaking more advanced games.

Using new Windows features wouldn't help developers make more advanced games.

btw u rly shouldnt mention idiocy. have you forgotten your entire history of posts during the introduction of the geforce fx? had the forums not been erased a few months back id have some fun digging up your old posts. you are quite possibly one of the dumbest individuals ive ever seen

Heh, go for it. The forums weren't erased.

they dont have to affect games, they offer the user more features that enhance the experience of windows

 

[Ailuros]

your whole train of thought is flawed.

Coming from you I consider it actually a compliment.

you support developers coding for the lowest hardware, yet you expect them to suddenly change when wgf is introduced?

I don't support anything in the nonsensical content you try to put it, just to get your even more unrealistic and nonsensical points across. I face reality as it is in order to get to more viable conclusions.

that will be even more advnaced then what we have now that is being wasted. 2005 will be no different than 2004. graphics will remain stagnent, as nothing in the forseeable future offers any step foward.

As I said buy yourself a console and call it a day, if that will really supposedly fix your problems.

 

[Scali]

This does have problems with normal mapping, specifically that tangeant-space normal mapping (required for most compression schemes, and often best for animated geometry) is invalid if the model is changed.

That is not a problem if you generate the high quality texture with the highpoly model and the low quality texture with the lowpoly model.

But still, games have had LOD for geometry for quite a long time. As an example, the Unreal engine has supported geometry LOD since around the days of the original UT. City of Heroes is another game that supports geometry LOD. It's just that it's typically used mostly to scale back the geometry of objects that are far away, not so much to reduce polycount on low-end hardware (however, it would pretty much do that automatically anyway, since lower resolution would naturally mean lower geometry LOD).

Yes, some games used geometry LOD, but it seems that it's gone out of fashion in the latest games. Doom3 only uses 1 LOD for example. And afaik, HL2 and FarCry do the same.
And as you say, even games with LOD generally didn't have a max-LOD level that could be set to best suit the hardware the game was running on.

 

[Laa-Yosh]

Like a set of lowres and highres textures, a game could also include a set of lowres and highres meshes, catering for various levels of 3d hardware.
With a modern game like Doom3 this takes very little effort, since the normalmapping already requires the artist to derive a lowres model from the original. He would just have to create two or more resolutions then. The tools for this are already available.

You're basically asking the modeling/rigging artists to perform 3 times the work.
But this is quite simple math: using the same budget, you can either have a range of art assets with one poly count, or you can have 3 different versions, but only 1/3 of the assets.
Now give us a reason why any sane developer would chose to have 1/3 of the content.

 

[Scali]

You're basically asking the modeling/rigging artists to perform 3 times the work.
But this is quite simple math: using the same budget, you can either have a range of art assets with one poly count, or you can have 3 different versions, but only 1/3 of the assets.
Now give us a reason why any sane developer would chose to have 1/3 of the content.

I disagree. The creation of the original high-poly character and textures and animation is by far the most time-consuming process. Creating the low-poly mesh and the normalmap is done by tools.
A subdivision surface modeler can generate low-poly versions of your model at the touch of a button. The normalmap is generated with a simple raycasting tool. If you know any modeler who can model, texture and animate as quickly as a modeler can generate a low-poly mesh and the raycasting tool can generate a normalmap, I'd sure like to see him at work

 

[KimB]

That is not a problem if you generate the high quality texture with the highpoly model and the low quality texture with the lowpoly model.

Sure, but that's a lot of data to have sitting around. Games already use gigabytes worth of data for geometry and textures. Some developers may consider this too much of a tradeoff.

You could, of course, store all normal maps in object space on the hard drive, and convert them appropriately depending upon the model at level load time, but with the massive amounts of data that games need to load, I don't know if it'd be that great to add another burden on load times.

Yes, some games used geometry LOD, but it seems that it's gone out of fashion in the latest games. Doom3 only uses 1 LOD for example. And afaik, HL2 and FarCry do the same.
And as you say, even games with LOD generally didn't have a max-LOD level that could be set to best suit the hardware the game was running on.

Well, one primary reason for this is simply that modern GPU's often perform best when there is no dynamic LOD.

I guess what I'm saying is that there are nontrivial problems to the greater adoption of geometry LOD techniques, so I don't expect them to really catch on until we get some robust HOS algorithms.

 

[KimB]

I disagree. The creation of the original high-poly character and textures and animation is by far the most time-consuming process. Creating the low-poly mesh and the normalmap is done by tools.

I don't think it's as simple as that. Automatic tools wouldn't produce models that would look very good. The normal mapping would be easy, of course, but I think you'd need to do a fair amount of tweaking to many lower-poly models.

 

[hovz]

chalnoth quotes

"The GeForce FX will automatically have less of a performance hit with AF just because it only has one texture unit per pixel pipeline."


"the simple fact is that the FX architecture isn't inherently slower than ATI's. Microsoft's spec is holding it back, which will mean that the FX's lower than the 5900 will need to sacrifice quality for any kind of speed (with no integer format, the drivers cannot possibly detect when integer format can be safely used, and it will have to be used anyway for speed).

Other than this, the FX line can be very fast, faster than ATI's in many cases, through optimization. And the FX line's shader performance is very, very complex. One cannot make any simple statement about it, except to look at peak performance and say it's harder to get that peak performance."


"First of all, single-texturing wouldn't be any better on an 8x1 architecture anyway, due to memory bandwidth limitations"


"For example if we, just for fun, average over all of these performances, we'll get a very, very rough 94% of the performance of a hypothetical FX with 8 full pipelines."


"I'm not so sure. I do still think that this 256-bit bus thing will fade in time, just like the multi-chip boards of the past didn't last very long. It's not economical. I think that the signal noise in the 256-bit bus parts will prevent them from really getting that close to the frequency of their 128-bit bus counterparts, which should help to keep the 256-bit bus parts from really dominating.

Another item of importance is what we've seen nVidia state in the past, that the focus is shifting off of memory bandwidth and towards computational efficiency. That is, if NV30-style shaders indeed require more processing power compared to memory bandwidth, then there won't really be a need to move to a 256-bit bus."


"And why do you think the FX will be slow with more advanced pixel shaders?

While its performance will definitely depend on the shader used, the FX does have much higher fillrate, and even higher PS performance when 16-bit floats are used"


"Hmm, I didn't notice the FX benchmarks previously. They really look like something major is wrong. In particular, the fixed-function performance offers promise to the eventual programmability performance the FX will expose. But the performance really is too low to assume it's anything but driver issues we're seeing here.

But, it doesn't matter a huge amount right now, though I really do expect usable DX9 drivers by the time the cards are available."

your nothing but a nvidia fanboy who regurgitates their pr material.

 

[Scali]

I don't think it's as simple as that. Automatic tools wouldn't produce models that would look very good. The normal mapping would be easy, of course, but I think you'd need to do a fair amount of tweaking to many lower-poly models.

Ofcourse there's nothing that doesn't take any time at all. But I believe that in the grand scheme of modern-day game development, where you have 2-4 years development time, I think that the extra work required would be insignificant. Especially considering the added value.
I suppose you don't even have to process all models either. For example, a crate only needs 12 triangles. Some models are low-poly by nature, so they don't require extra LOD.
No, my artists have to just create some extra models with different LOD

 

[Laa-Yosh]

I disagree. The creation of the original high-poly character and textures and animation is by far the most time-consuming process. Creating the low-poly mesh and the normalmap is done by tools.

It is true that one can easily generate several sets of normal maps for different LODs of the same mesh. But the LOD meshes have to be created somehow, and no automatic tool can give us meshes that are good enough to be used as a fallback for lowend systems.
The low-poly mesh is only good if the polygons are distributed very carefully to 1. provide as much shiluette detail as possible 2. allow good skinning. Neither can be completed by an automatic tool; this is also the reason why Multires and other tools have only been used for the least detailed LOD in some games because it's only seen from very far away. The algorythms are simply not intelligent and thus cannot decide which vertices to keep to preserve the details and features of the mesh.

And keep in mind that the mesh has to be bound to the skeleton, and the rigging process can not be automated - someone has to set the bone assignments and weighting, thus it takes artist time, too. UVs are somewhat preserved, but need a varying amount of tweaking and of course thorough inspection.


(I decided to add that one of my tasks at my first job in 2001 has been creating LOD levels for characters using Multires and lots of manual cleanup in 3ds Max, so I have actual hands-on experience with the tools.)

A subdivision surface modeler can generate low-poly versions of your model at the touch of a button.

This is complete nonsense. Subdivision increases "detail", or rather, it tesselates curves to a higher degree, it smooths the model. Thus it cannot be used to generate lower polygon count models; you can only choose to have less iterations of the subdiv scheme, but that means more like "it does not increase the poly count as much". The most commonly used scheme BTW is the Catmull-Clark one, which quadruples the polycount with each iteration.