"Yes, but how many polygons?" An artist blog entry with interesting numbers

It's been said before but the issue with smoothing/rounding, is that the exact shape of the curve matters and often that means a lot of control points which means it can actually be cheaper just to add the polygons to the models.

There are obvious use cases for GPU Tessellation, terrain being perhaps the single most obvious one. And I do think we'll see more significant use of it as the generation progresses and it's better integrated into various art pipelines.

 

Yes but it's not enough, the curve just ends up looking wrong. Subdivision surfaces are the classic use a polygon mesh to drive a set of curves, but when I experimented with then 10years ago for car models, the control meshes to get the curves looking right were more dense than any model we'd ever actually display.

There has to be a win somewhere to make it worthwhile, displacement maps or similar might be it, but none of that stuff is in the pipeline at this point.

 

Thanks ERP. Wouldn't such a thing be useable in an application where the curves don't have to look as intended either? Just to make it look a bit richer? I mean per vertex gouraud shading looks better than flat shading too, despite of not being completely correct.

 

We've been dynamically tessellating terrain for LOD for a decade at least! Come on, technology. Go somewhere!

 

If there is value to it it'll get used.
I would assume it will be available in all the next gen consoles, so it's just a question of finding the value.

 

It'd introduce an additional and quite complicated step in the character pipeline which is complex enough already; and as I've mentioned, in Last of Us we've seen that even this hardware generation is capable of quite impressive results.

If you look at that game and ask what could be used to improve those characters, more polygons would be quite far down the list IMHO. Even better (more precise) lighting and shadows, more complex shaders, better hair, dynamic clothing are all more important to get a better result than more triangles, I'd say.

 

A great example would be kratos from god of war 3, i believe he was in the 25k range now compare him to bayonetta or ninja gaiden 3 characters which double him. Why is that?

 

Thanks. Looks like they bumped the model's complexity quite a bit from Witcher 1. I'm curious on how Witcher 3 will turn out.

 

Kratos doesn't has any "costume" apart from the loincloth, nor has he got any accessories. Also female characters usually end up having more polygons.

 

Exactly, accessories are where most of the polygons in game characters are spent, and of course boobz.

 

Kratos also did not have toes, and flat chains around his fists. This might sound pety but they always felt like pretty jarring and obvious shortcoming for such a great looking game.

 

A good example of that is Bayonetta, who has very fancy clothing, hair, and each of her four small guns are 8k polygon each, This boosts her character model to almost 54k polygons. It looks like Bayonetta 2 make even be beyond that.

 

He's also balled.....

If you look at momiji and bayonetta they have a lot of polys in their hair as well. Kratos in particular is what you would call an efficiently made character do to style and purpose. with that taken into account they can easily get to work on the textures and shaders.

Besides the polys if you look at the textures of a Doa/Ninja Gaiden character they look smooth (for their female characters anyways.) so that could give a sense of a plastic appearance, resembling somewhat of low detail. (possibly making it hard for some to believe the number in polys.)

 

Having a lot of polygons in the main character is actually not such a big thing, when you're already pushing millions of polygons per frame, it doesn't matter that much if just that one character is not 15K but 50K or even more. It also doesn't usually take up significant screen space, so even if the little triangles ruin the fragment shading efficiency, it doesn't hurt as much.

The reason we don't see it that often is probably that most studios want to keep their characters consistent, they don't want the player's avatar to stand out too much from the rest of the cast.

 

I've noticed playable characters more often looked better than NPCs during the PS1 and PS2 eras. A classic example of this is in the first three Spyro games. Spyro almost always looked better than anyone else.

Probably the only example off the top of my head in the current gen where the main character looked dramatically better than most NPCs is Prototype 1... But that's an open world game that had probably over 100 different civilian designs alone.

 

Uncharted 2-3 has considerably more detail in the main character, and we also had some exact numbers on this.

With other games it's usually just more refinement and care, but significantly higher poly budgets are indeed rare. However in cutscene heavy games the leads usually have higher texture resolution on the areas that are most visible (head and upper torso).

 

in the ps4 announcement trailer David cage said

Omikron: The Nomad Soul 150 poly

Fahrenheit 1500

Heavy Rain 15000

Kara tech demo 25000

Beyond: Two Souls 30000

they also had a amazing tech demo of an old man head like the 1999 ps2 square old man that looked very real but he gave no numbers

 

The head looked like it had a moderate poly count and normal maps, no sign of tessellation and/or displacement. But it did look quite nice.

 

damn, you beat me to it!

Heavy Rain's character's looked pretty nice for 15k!

PS4 was announced, and to my prediction it's going to support plenty of DX11 features. an X86 based APU and 8 GBs of GDDR5 unified memory, plus a near 2 Tflops GPU!

.......I don't think i should be able to explain how much polygons it should be able to draw.

 

50 to 100k can be expected for main characters.