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

Where'd you get that from? The game models don't look half as good according to my eyes.

from Kojima productions via a XSI case study -

"Most of the characters that are animated on the console, including the main character, Snake, have been restricted to a data size (including the face model) of about 5,000 to 10,000 polygons. Further, characters are used that have the same polygon resolution in both the game action and the event demos."


Friendly Mod Note: The link to the article was posted a page back - http://forum.beyond3d.com/showpost.php?p=1217466&postcount=443
-AlS
 
from Kojima productions via a XSI case study -

"Most of the characters that are animated on the console, including the main character, Snake, have been restricted to a data size (including the face model) of about 5,000 to 10,000 polygons. Further, characters are used that have the same polygon resolution in both the game action and the event demos."

That pretty much settles it. Amazing when I think about those being furious about 15-20k polygon estimation for Snake done by someone that got under fire for saying that as a professional modeler! :LOL:
 
That pretty much settles it. Amazing when I think about those being furious about 15-20k polygon estimation for Snake done by someone that got under fire for saying that as a professional modeler! :LOL:

I thought that making a game look good with fewer polygons was a good thing, so the whole discussion was a bit confusing to begin with.
 
from Kojima productions via a XSI case study -

"Most of the characters that are animated on the console, including the main character, Snake, have been restricted to a data size (including the face model) of about 5,000 to 10,000 polygons. Further, characters are used that have the same polygon resolution in both the game action and the event demos."


Friendly Mod Note: The link to the article was posted a page back - http://forum.beyond3d.com/showpost.php?p=1217466&postcount=443
-AlS
event demos = cut scenes?
 
Which game has the record in total number of polys per frame?
That's a trick question. Do you mean visible, rendered polygons, most polygons in the models without regard for culling etc., overall triangles used including invisible ones use ion effects, or what? It's actually a pretty stupid metric is you think about it. There are a million pixels in a 720p frame. You could have one triangle for each pixel and that would only require 1 million polygons. The use of > 1 million triangles is only because we can't yet optimize our polygon usage effectively. A game that did achieve this could get far better results from a much smaller polygon number than a typical engine, making the comparison with other titles pushing more polygons asinine.
 
That's a trick question. Do you mean visible, rendered polygons, most polygons in the models without regard for culling etc., overall triangles used including invisible ones use ion effects, or what? It's actually a pretty stupid metric is you think about it. There are a million pixels in a 720p frame. You could have one triangle for each pixel and that would only require 1 million polygons. The use of > 1 million triangles is only because we can't yet optimize our polygon usage effectively. A game that did achieve this could get far better results from a much smaller polygon number than a typical engine, making the comparison with other titles pushing more polygons asinine.

Thx for explain, very interesting.

I remember Rogue Squadron II from Game Cube using a total of 6 million of polys with various texture layers, right?

A game of today have to use almost this number x 10.
 
the number of triangles rendered usually aint the number of triangles displayed
eg
depth pass
color pass
shadowmap passes
other effects
not to mention backface culling + clipping

typically the number of triangles visable are gonna be <1/3 of those drawn
 
Also remember that vertex to triangle ratio is highly dependent of the content. For example every change in smoothing groups, materials, UVs or other per-vertex values causes the hardware to split the edges and draw the mesh in separate parts, doubling the vertex count on these edges. And there's no way to avoid UV seams on complex shapes so there's no way to completely avoid this.
 
Thx for explain, very interesting.

I remember Rogue Squadron II from Game Cube using a total of 6 million of polys with various texture layers, right?

A game of today have to use almost this number x 10.
Unless I'm mistaken I think it used closer to 12 Million :)
 
Unless I'm mistaken I think it used closer to 12 Million :)

But that is per second, not per frame. to get per frame you need to divide that sum with amount of frames per second to get X/polygons per frame.

if 12m per second then...

@ 30fps it's 400k/polygons per frame
@ 60fps it's 200k/polygons per frame
 
But that is per second, not per frame. to get per frame you need to divide that sum with amount of frames per second to get X/polygons per frame.

if 12m per second then...

@ 30fps it's 400k/polygons per frame
@ 60fps it's 200k/polygons per frame
Ok thanks for the correction, I slightly misread the post.
 
from Kojima productions via a XSI case study -

"Most of the characters that are animated on the console, including the main character, Snake, have been restricted to a data size (including the face model) of about 5,000 to 10,000 polygons. Further, characters are used that have the same polygon resolution in both the game action and the event demos."
There's just no way in hell (pardon my French) that that can be true. The cinematic/cutscene models (which you can observe very closely in the camouflage viewer screen) have all sorts of special effects and little details, such as wrinkle maps, and textile/fabric textures, that just aren't present in the in-game model. I mean, you can see for yourself when you're installing MGS4 or installing a new chapter, when Snake takes a smoke break, you can see the normal maps and bump maps and wrinkles change on the side of his face and cheek/jaw bones while he smokes. You don't have anywhere near that level of detail on Snake's face when you're playing the game and controlling the camera to look closely at his in-game model.

Not to mention that the level of expressiveness is just not there on the face of the in-game model. Zoom up the camera to Snake's face, and you see a pretty standard, expressionless robot face. An expression that hardly changes. If what's been said is true, then that would just be a monumental waste of the polygon budget, to have that much potential for facial expressiveness, but not use it (or have the need to use it) at all. There's no reason to have an in-game polygon model that has the same number of polygons as the cutscenes; it just doesn't make sense, and would only be an unnecessary drain on performance.

I mean, the polygon count difference is a little less certain than the clear, clear differences in texture quality. But all the same, it's very doubtful that the models/meshes for the in-game and cinematics versions of Snake are one and the same. I don't think textures are everything that account for the difference in quality.

I smell either PR exaggerations, or misunderstandings in the translation.
 
Last edited by a moderator:
the loading screen char is a quite detailed, but so he should...he's face is taking up half the screen...so no doubt the res might be double that of in-game.

but I also dont doubt the article at all, as im working with konami on a project now (albiet, not kojima productions) and the texture / poly budgets are quite similar (we are not using things like gradient spec maps etc though)
 
...The cinematic/cutscene models (which you can observe very closely in the camouflage viewer screen) have all sorts of special effects and little details, such as wrinkle maps, and textile/fabric textures, that just aren't present in the in-game model.
Doesn't the article say only the poly counts are the same? Certainly you're quoted passage doesn't mention anything about facial details being identical. Changing detail maps out would be an easier way to accomodate memory changes between gameplay and cutscenes.
 
But that is per second, not per frame. to get per frame you need to divide that sum with amount of frames per second to get X/polygons per frame.

if 12m per second then...

@ 30fps it's 400k/polygons per frame
@ 60fps it's 200k/polygons per frame

Actually they claimed 16-18 million polys per second for the successor (RS III: Rebel Strike), but as Shifty pointed out earlier that includes culled polys, etc.
 
Back
Top