hasanahmad
Banned
SanGreal said:
the article didnt point to ALUs not wasting time. its that the ALUs performance is 90% of theoretical performance achievable rather than 50-60% in current high def graphic cards
How many Polys can the New Consoles Push?
- Thread starter BenQ
- Start date
the article didnt point to ALUs not wasting time. its that the ALUs performance is 90% of theoretical performance achievable rather than 50-60% in current high def graphic cards
hasanahmad
Banned
im confused because look at this chart
http://www.gentechpc.com/products/Build/vga.htm
it has a Geforce4 TI 4400 having the same amount of vertices/s as Xbox GPu which had a Geforce3 which leads me the believe when Nvidia uses the word vertices it doesnt mean triangles per second
http://www.gentechpc.com/products/Build/vga.htm
it has a Geforce4 TI 4400 having the same amount of vertices/s as Xbox GPu which had a Geforce3 which leads me the believe when Nvidia uses the word vertices it doesnt mean triangles per second
hasanahmad said:
The reason it is able to reach a higher percentage of its theoretical performance is because ALUs will spend less time being idle. With a traditional architecture your vertex shaders could be sitting there doing nothing waiting for work to finish on the pixel shaders. The point of a unified architecture is to prevent things like that from happening.
I'm still not sure why we're on this tangent about the USA though.
hasanahmad said:
It could, as Titanio mentioned on the first page, be referring to the vertex transform rate and not the setup rate I suppose?
Perhaps I'm just in over my head on this debate (as I am quite the novice when it comes to these things), because I'm not sure what you are even trying to suggest.
The efficiency drops when you can't supply the ALUs with info to work on. In the case of separate shaders if you have 6 vertex shaders and 24 pixel shaders, if you have no vertices to work on, 25% of you ALU's are sitting idle. Whereas on a unifed shader system if there's no vertex work to be done the ALU's can pick up pixel work. That's the place the US system gets it's improvement in efficiency. There's no difference in how it constructs or renders trianlge to cause more efficient operation. AFAIK there's no problems with triangle setup or area to make improvementshasanahmad said:
hasanahmad
Banned
i think its transforms per second because X850 has 600 million transforms per second which is the same as 6800 (600 million vertices per second)
now tell me whats the difference between transform rate and setup rate
now tell me whats the difference between transform rate and setup rate
hasanahmad said:
I may be wrong, my memory is somewhat hazy on this, but the Xbox GPU I believe has twice the number of vertex shaders as the Geforce3, and the same as the Geforce4, hence the same figures?
Anyway, I'm not sure it's certain as to whether that's the transform or setup rate.
As for whether X360 has more potential to reach its peak..a system being XX% efficient does not mean it will come within XX% of its peaks at all times. It depends how much work you're doing on each vertex, and also on your pixels. PS2 could probably reach its peak just drawing unshaded triangles on the screen, as much as X360 could reach its peak probably with very simple shaders (I believe a dev here once said he thought it might reach its peak with Xbox1-level shaders?).
hasanahmad said:
Transformation is basically moving vertices about. You mutliply your vertices by a matrix to translate, rotate, shear, scale etc. your vertices. The peak rate is how many of these you could do with the simplest of transformations, perhaps when using some other tricks to boot.
IIRC, setup happens just before rasterisation (filling in the pixels)..it's basically calculating slope and delta values for colour/depth between the vertices of the triangle, to set things up for interpolation during rasterisation.
Last edited by a moderator:
Oh for god's sake stop arguing such elementary stuff. Any closed mesh without parts sticking out will have a similar number of vertices and polygons.
And even if it has some parts sticking out, the number of vertices will NEVER be 3 times as many as the numbers of polygons, not even close.
Number of vertices will only be 3 times as many as the number of polygons if each and every triangle is detached from each other, and that means only cases with polygons flying around (after Burnout-style crashes) or smoke and fire effects using alpha textured polygons. If we start using particles for smoke or fire effects, that's back to 1 vertex each.
And even if it has some parts sticking out, the number of vertices will NEVER be 3 times as many as the numbers of polygons, not even close.
Number of vertices will only be 3 times as many as the number of polygons if each and every triangle is detached from each other, and that means only cases with polygons flying around (after Burnout-style crashes) or smoke and fire effects using alpha textured polygons. If we start using particles for smoke or fire effects, that's back to 1 vertex each.
DotProduct
Newcomer
Actually it is quite easy to derive that the number of polygons is about twice the number of vertices for every manifold 3D mesh.
In praxis you experince the same ratio.
Euler´s Formula for genus 0: V-E+F=2
(The actual proof is an easy exercise
)
In praxis you experince the same ratio.
Euler´s Formula for genus 0: V-E+F=2
(The actual proof is an easy exercise
)DotProduct said:Actually it is quite easy to derive that the number of polygons is about twice the number of vertices for every manifold 3D mesh.
In praxis you experince the same ratio.
Euler´s Formula for genus 0: V-E+F=2
(The actual proof is an easy exercise
So your saying that the number of poly's to vertices is "about" 2:1
So the Ps3 can push 1.1 billion vertices / second = 550 million poly's / second
and the Xbox 360 can push 500 million poly's / second = 1 Billion vertices / second.
Is that what your saying?
And your "easy excersise" is a joke! I don't get it AT ALL
Last edited by a moderator:
Polys = 2x vertices.
ergo 1.1 billion verts = 2.2 billion tri's.
That's going by dotproduct's assertion. I don't know how true this is. But I'd certainly accept 1 billion tri's a second, but again, that's if the 1.1 billion figure is right...
Oh, and the Euler equation is probably very easy to mathematically prove if you're good at maths, but most people (me included!) aren't!
Shifty Geezer said:Polys = 2x vertices.
ergo 1.1 billion verts = 2.2 billion tri's.
That's going by dotproduct's assertion. I don't know how true this is. But I'd certainly accept 1 billion tri's a second, but again, that's if the 1.1 billion figure is right...
Oh, and the Euler equation is probably very easy to mathematically prove if you're good at maths, but most people (me included!) aren't!
no no no, there will never be more polygons than verts. that's impossible.
Shifty Geezer said:Polys = 2x vertices.
ergo 1.1 billion verts = 2.2 billion tri's.
That's going by dotproduct's assertion. I don't know how true this is. But I'd certainly accept 1 billion tri's a second, but again, that's if the 1.1 billion figure is right...
Oh, and the Euler equation is probably very easy to mathematically prove if you're good at maths, but most people (me included!) aren't!
No.
They are going to list the spec that makes it look the best. They aren't going to list it as 1.1 billion vertices if they can claim 2.2 billion poly's.
You don't really believe the PS3 can push 2.2 Billion poly's / second do you?
No. I don't agree with dotproduct's assertion. I think vertex : poly's is around 1:1, so it'll be 1.1 billion tris. And I don't know how reliable that figure is either, but as a peak figure it's not outside the realm of possibility. That's 20x the PS2's peak which given advances in tech seem reasonable.
Shifty Geezer said:
Why don't you agree with him? You were so QUICK to agree with him when you thought that 1.1 B vertices would = 2.2 B tri's.....hmmmmm....
There has been talk in this thread about a 1:1 ratio for polys to vertices but those have only been under rare circumstances such as rendering a sphere or pyrimid.
Oh and if Dot Product would like to explain that mathematical rubixcube he posted.....that'd be good = P
robofunk said:
Well it's wrong.
Similar threads
