Setup Limit of PC GPUs

[pjbliverpool]

I was wondering if anyone had any confirmed information on the triangle setup limit of modern PC GPU's.

We know Xenos is 1 per cycle and RSX has been confirmed as 0.5 (I think).

Does this mean all G70/71's are only 0.5 triangles per cycle?

What about ATI GPU's?

According to Dave's article here:

http://www.beyond3d.com/reviews/ati/r420_x800/index.php?p=6

R420 can output 1.5 triangles pr clock but is this simply referring to vertice transforms?

Is this something we can get cleared up?

 

[Dave B(TotalVR)]

Well, with 1.5 triangles per clock and a 600Mhz clock speed thats a theoretical limit of 900 million triangles per second.

are we anywhere near to that speed in VS performance yet?!?!?

nope...

 

[pjbliverpool]

Well, with 1.5 triangles per clock and a 600Mhz clock speed thats a theoretical limit of 900 million triangles per second.

are we anywhere near to that speed in VS performance yet?!?!?

nope...

I suspect though that 1.5 referred to vertice transforms rather than triangles setup. 6 vertex shaders would indeed equal 1.5 vertice transforms per clock (I think), and he also mentions that its up 50% from the R380 which only had 4 vertex shaders.

Im wondering what the R580 setup limit is though. Is it also 1 per clock like Xenos or is it in line with G70, i.e. 1/2 per clock.

 

[Zeross]

1.5 vertex / cycle is the peak transform rate on the R420, 2 vertex / cycle is the peak transform rate on R520/R580 and G70. I was told by ATI at the R520 launch that the setup rate for the R520 was 1 triangle / cycle, I guess it hadn't changed with the R580.

 

[Jawed]

1.5 vertex / cycle is the peak transform rate on the R420, 2 vertex / cycle is the peak transform rate on R520/R580 and G70. I was told by ATI at the R520 launch that the setup rate for the R520 was 1 triangle / cycle, I guess it hadn't changed with the R580.

The data received by the vertex processing units includes three-dimensional position information, texture co-ordinates, normal vectors, colors, and material information. The most basic function of these units is to perform transformation operations on the position data, which typically consists of four 32-bit values for each vertex (x, y & z co-ordinates, plus a w co-ordinate used for calculating perspective). This type of transformation requires a 4x4 matrix multiplication, which can be broken down into four dot product instructions. Each 128-bit vector ALU in the RADEON X800 is designed to execute one of these dot product instructions every clock cycle. Since there are six of these ALUs, a total of 1.5 vertex transformations can be processed every clock cycle.

Vertex shaders normally do more than simply transform vertices though:

http://www.pieterg.com/Tutorials/hlsl4.php

Jawed

 

[pjbliverpool]

1.5 vertex / cycle is the peak transform rate on the R420, 2 vertex / cycle is the peak transform rate on R520/R580 and G70. I was told by ATI at the R520 launch that the setup rate for the R520 was 1 triangle / cycle, I guess it hadn't changed with the R580.

Thanks for that, its just the kind of information I was looking for. I assume the G7x parts on the PC have only half that given the recent news about RSX. Funny that ATI doesn't make more of an issue of that particular spec. I know its not too relevant in the real world but most people pay a lot of attention to polygon counts and would draw big conclusions from ATI's top part pushing double that of Nvidia's.