Are triangle counts a thing of the past?

MeshMagnet

With the new consoles and high end video cards coming out on a yearly basis; would you say that triangle counts no longer matter?

I've heard from various conversations that most consoles and high end PC's can easily push over a billion poly's on the screen. Would you say that is true?

What would be the limit for say a character in a FPS game on the new consoles?
1 million tri's? Or is that a bit overboard?

What about a vehicle like a tank on the new consoles? Would 2-5 million tri's be a bit much?

ZacD

That is way overboard. 40k tri for a high end character or vehicle is acceptable.

MeshMagnet

That seems kinda low TBH. How are you getting these numbers?

I've heard though in the game The Division, the cars had something like a million tris. Can anybody verify that?

ZacD

I'm going off Ryze and Killzone Shadowfall, the 2 of the bigger and better looking next gen launch titles.

Car games are a whole other thing, Forza might have million poly cars.

Optinium

Unshaded triangles on screen could probably be in the billions yes, is that going to happen in games, probably not.

A million tri's in cars on an open world online multiplayer? Hmmm I'd say that's bogus but then again I'm not working on it, I'd expect cars in a racing game (say the next Gran Turismo) to be hitting figures upwards of that but that's a bit more contained.

40K as ZacD mention is a figure that's used in titles at the moment (KillZone:SF, Beyond Two Souls etc...) I'd expect close up characters to go upto 100k or thereabouts, there is a point you reach when actual geometry @ 1080 makes little difference.

Froyok

MeshMagnet wrote: »

I've heard though in the game The Division, the cars had something like a million tris. Can anybody verify that?

1 million ? Really ? -_o
I have serious doubts about this number.

Even characters don't go this far while they need micro-animations to be believable. So a car in a game like Division... well... That sounds very overkill.

MeshMagnet

I'm delving a bit deeper and found that some of the characters in Uncharted for the PS3 where hitting 35k. That's on a 10 year old console; right?

Shouldn't we be able to push that way beyond that now?

Source:
http://forum.beyond3d.com/showthread.php?t=43975


Uncharted: Drake's Fortune, PS3, 2007
Main characters - ~20,000-30,000 polygons
Drake - ~30,000 polygons
Pirates - ~12,000-15,000 polygons

Uncharted 2
Chloe - 45,000, each subsequent LOD has halved count (3 LODs)
Drake - 37,000, no LOD in SP
Drake's Hair - 4,000

Optinium

Counts will vary widely depending on game type, engine and the tools used to produce it (eg a million fully shaded tri's in Maya starts to chug big time) etc... And as I said earlier once you get to a certain point, geometry doesn't matter, it just becomes unruly and harder to manipulate (40k tri's is easier to work with then a million).

As an example you don't need a million tri's to define a perfect looking sphere at mid distance @ 1080, a couple of thousand, if that, would probably do.

So can consoles push more? Yes ofcourse.
Do they need to? Nope

Bigger textures, better animation and more FX are likely to make a bigger visual impact then more triangles after a certain point.

ZacD

For in-game characters, the PS3 game used three different LOD models (more polygons used the closer you are to the character in question), up to 10,000 polygons and 1024x1024 textures. Things have changed significantly for PlayStation 4 with seven LOD models and a maximum of 40,000 polygons and up to six 2048x2048 textures.

At 1080p characters don't look much better after 40k tris. Instead of pushing for a higher polycount, engines are pushing for more accurate shader and lighting models, higher resolution textures, better reflections, effects, etc.

The Flying Monk

Don't forget that mobile still exists as a platform. Even a high end Iphone is still not as powerfull as an xbox360/ps3 in most areas. And not every game is made for a high end PC or the latest console.

And for a game like Uncharted or The Last of Us, the polycount could also be limited to the complexity of the characters rig.
The more vertices and more bones make a lot more transformations you have to do per character.

MeshMagnet

Okay; would you kick me out of bed if I gave you a battleship model that consisted of 3 million tris for an XBone or PS4 game? Or would I just be crazy?

stevston89

Yes. I wouldn't even consider it. Not only would it be hard to get the performance good, but it just shows you can't optimize properly. You could do the same thing with an fraction of the tris. The point is you just don't need it unless you plan on being able to zoom into every nook and cranny of the ship. It's better to make something with only the tris you need than throw tris at it for no reason.

Popeye9

Here is a comparison of Triangle count.

http://wccftech.com/ryse-polygon-count-comparision-aaa-titles-crysis-star-citizen/

The Flying Monk

Is this battleship going to be something I can run around on in first person? Or is it going to be in a scene with 100 other ships? It's not just about proformance. If your making a 3 mil tri object that doesn't need that much detail, then your wasting time.
Or maybe your not working on a big AAA epic game and only have a short amount of time to make assets.
Just because you can thow a millions of tri's at your scene, doesn't meen you should or have to.

Ultimatly it all depends on multiple factors like platform, contex and time and budget constraints. It will be up to the lead artists and programers to give you a mesh budget.
If your lead tells you something has to be under 10,000 tris, then you'd better be able to do that. Or have a good justifation for why its not.

MeshMagnet

I was thinking something like an FPS where you'd be able to run around the whole ship.

ZacD

Making a level/map is completely different than making one object.

MeshMagnet

Found a great paper about this; PS4's Killzone Postmortem:

http://www.guerrilla-games.com/presentations/Valient_Killzone_Shadow_Fall_Demo_Postmortem.pdf

Yeah, it looks like I was spouting out some insane numbers.

stevston89

Just to clarify as well if it is supposed to be an entire level for an FPS. 3 million tris doesn't seem insane, but you wouldn't build it as on piece and you certainly wouldn't be drawing them all at the same time. You would hide the stuff you can't see to save cost and to do that you would want to break it up into pieces.

Neox

well with tesselation and displacement you could get your mesh to eat up a whole bunch of polygons very fast, however ever the tesselated mesh would still be pretty simple. and be it for rigging and skinning purposes. a fully modelled character will rarely ever have 1 million unique triangles, subdivision comes to mind to smooth out lowpoly silhouettes, but even then how close do you want to go? A million is a whole lot of polygons, you better have a lot of hair and fur going on to really demand that kind of detail ^^

Mark Dygert

I think the old rules of...

"use what you need to make it look good, no more, no less."
"Waste as little as possible and use those resources where they matter"

...still apply.

I think tri-counts will always matter but they might stop mattering to artists at some point in the future. By that I don't mean tri counts so high it feels unlimited, but I think we are closer to "seamless, automated, scalability" than we've been in the past.

We are getting closer to throwing a million poly model at an engine and it uses what it needs, when it needs it.

Can your model be down res'ed to run on mobile and up res'ed to run at cinematic quality? Soon I don't think we'll be worried about making each level of detail as we've been in the past and I don't think we'll be throwing the highest sub-d out of zbrush into game engines, but making models that transition and scale smoothly as needed.

There are already a lot of tools that optimize models on the fly, adding tessellation and other techniques will get us closer. But we'll probably need some innovation on the art front to get over the hurdle. We need to allow artists to see all the levels and fine tune them as needed. Not just to get static details right but also for bone structures, skin weighting and deformation.

How you rig and weight an ultra low poly model isn't the same as an ultra high, so not only does detail need to scale, but so do rigs and animations, that's a bit trickier.

We're still pretty early in the next generation and doing things pretty much the same old way with higher limits, but we have a lot of resources and just throwing them all at tris or bigger texture sheets might not be the best way to make some games better.

Personally this question seems like someone who is used to zBrush hoping to get out of learning deformation and lower poly modeling techniques, ha. I don't think that's going away anytime soon...