If X360 capable of 500 million polygons/sec. How to explain?

[Love_In_Rio]

Pure especulation:

Do you think it will be its maximun capacity using the 48 shaders units of R500, or using for example half of them and this way being able to be treated in the other free shaders for pixeling? or on the other hand do you think this figure could be reached using as much VMX as Shader units ?

I supposse they will be flat polygons anyway. So if Xbox was capable of 125 millions ( i dont think this was real ) then Xbox360 will have 4 times the rendering capacity, won´t it ?

 

[Acert93]

My guess is we need more information, but...

http://www.beyond3d.com/forum/viewtopic.php?t=22753&start=60

Last years leak said these were real stats for non trivial shaders. The question we all had was when these stats were released it gave something like 10x the vertex shader power and 4x pixel shader power of the X800--so the question was were these comparisons if the chip was ONLY doing either/or, but not both at the same time. One of the patents indicated that the CPU can assist with vertex load... it will be interesting to see how this really performs, but my guess based on the TeamXbox leak that those are fairly realistic if not too narrow. Game design (poly heavy or texture heavy) will probably play a big role in how it is utilized.

As for the Xbox comparison, it is hard to compare. Even if it was only a 4x fold increase in raw poly pushing power, the reality is GPUs have been focusing on ways to make what they have go farther. In this regards modern GPUs outperform in featureset by a large gap. So the question is not just the mere poly power, but what features it can have at such a benchmark. If the Xbox 360 can do 500M poly/s with effects on vs. XBox 1's 125M flat shaded, then there is no real comparison. Even Nintendo's 10-12M in game number does not compare because the new consoles will have more features available.

Sort of apples-to-oranges without having more information. That is why those non-detailed numbers were a waste to begin with. Featureset and performance with those features is more important than just raw numbers.

 

[version]

Xbox360 real specs:

3* 3.2 Ghz ppc core
ati gpu 500 Mhz with 10 MB eDram
500 million polygons/s
48 billion shaderoperations/s
512 MB gddr3

i have 3 big real scans

 

[ERP]

Because polygon pushing power isn't just limited the ALU speed.

It happens that on Xbox that triangle setup is fast enough that other than totally trivial shaders it's never a bottleneck (not quite always true as it happens).

This will likely not be the case for nextgen systems or graphics boards.

As chips get faster the trend is towards more ops per pixel/vertex, so having triangle setup that can match your peak vertex transform rate is just a waste of die space.

 

[BOOMEXPLODE]

It could be that the 500 million number is assuming one or more of the CPU cores is helping out with T&L. Maybe the CPU does the transform, the GPU does the lighting.

 

[nAo]

It's simple: R500 can't setup more than one primitive per clock, that's all
If the hw is executing a simple (vertex and pixel) shader and thus it can produce more than 1 shaded vertex per clock (I'm assuming a one vertex <-> one primitive case) ALUs will stall.