What's behind the apparent need for more polys in many next-gen games?

[zidane1strife]

From what I've heard the next-gen consoles should be able to pump out more polys then you could ever wish for. Thus the question: why are we still seeing rough areas even when using inanimate objects?

kameo, Rumble Rose sequel come to mind, but given this are basically ports(still why did RR sequel switched to 2d crowds?), it may be reasonable.

But those are not all, even some of the very best of next-gen games appear to've such limitations when you get up close.

So what gives? Is it modelling-time constraints? first-gen expertise constraints? tech constraints?

 

[Powderkeg]

2 main factors.

#1 is always $$$$$. It takes time to program that much geomitry. Time is money. The question is, does spending the extra time to generate the extra geomitry generate any monitary return. If the answer is no, they don't do it.

#2. Hardware limitations. Until we reach the day where 1 pixel = 1 polygon, there will always be straight edges, and visible "rough" areas.

 

[Shifty Geezer]

I think the GPU's can handle more poly's than pixels, but the rest of the system is the bottleneck. All those models, textures, shaders, and whatnot is more than the non-triangle processing can keep up with compared to poly counts.

 

[Squeak]

Bandwidth, is the main limitation I would imagine.
Vertices take up much more memory and bandwidth than pixels/texels.

Only Cell with it's large amount of ondie memory, could possibly do some kind of tile rendering in the SPUs.
The result would be send to the shared VRAM with Z-buffer data, to be blended with the GPUs part of the scene.

 

[jvd]

Major reasons ?


These are launch titles and thus are

1) Ports from the old systems

2) games that were designed on the old systems and moved late in development to the new systems (kameo , pdz )

3) just crappy games of which we have all generations


I thinking looking out a year from launch on both systems the minimum quality is going to advance to pgr 3 / gow lvls and the games that look like high res xbox games will be gone . Its just part of the transition

 

[randycat99]

I throw my guess in as memory capacity. Shuffling fine-poly models around can eat up memory quickly, imo. So it's a fine balance between your game code, all of your textures and sound effects, and sheer model density, while managing not to hit that last MB of memory punctuated by a spectacular game crash.

 

[Java_man]

Bandwidth, is the main limitation I would imagine.
Vertices take up much more memory and bandwidth than pixels/texels.

This make me wonder why console makers don't spend much more ressources on memory speed,latency and architecture.
I would really to see one day a console where an important part of the budget would be spent on the memory architecture(perharps even development of new kind of memories) and less money would be spent on the CPU and/or GPU(besides of course to better take advantage of this memory arcitecture).

 

[V3]

Insufficient LOD is my guess.

 

[Guden Oden]

Probably all of the already mentioned reasons, plus a couple other ones. Development software is either inadequate or improperly used; if a character is modelled using higher order patches, it can be scaled in polycount freely, but some devs undoubtedly model their models directly in polygons instead, making it much harder to scale. After all, a program can't know when a straight corner in a model should be a straight corner or a rounded one.

John Carmack said around the quake3 generation that each vertex occupies around 30 bytes of memory, today in the DX9 age it might be even more. Xenos is said to have a poly setup limit of 500M/s, and assuming a rough 1 vertex/poly ratio, that's 8.33M vertices per frame at 60Hz. That's nearly 240MB of memory's worth of vertices! Clearly devs can't sacrifice nearly half the main memory of the console for one frame's worth of polys. It might work going overboard in a fighting game with one extremely detailed arena, but in a first/third person game you'd need a large sprawling level to explore, and it doesn't work to use up half the memory on one single scene.

Clearly, procedural geometry has to be heavily exploited before we can even come close of scratching the theoretical limits of xenos. Question is, when will this happen in the console's lifespan? Not in the launch games, and likely not in the entire first generation other than in peripheral applications I'd think.

Maybe if x360 has a longer life than the original box, we'll see some good use in the later generations of software, when devs know how to split up work between the three CPU cores and make things run smoothly.