Halo 3 runs at 640p native XD

[makingmusic476]

Entroper said:

makingmusic476 said:

You can't just compare the number of pipelines between the two GPUs. The RSX only has 24 pixel shaders and 8 vertex shaders, compared to the 360's 48 unified shaders, but an individual unified shader is not as good as a dedicated shader for any specific task. The RSX's vertex shaders can handle 5 ops per cycle and the pixel shaders can handle 2 ops per sec, whereas all of the Xenos' shaders are only capable of 2 ops per second.

This leads to a a maximum of 136 ops per cycle for the RSX ((24 x 5) + (8 x 2) = 136) compared to a maximum of 96 ops per cycle for the Xenos (48 x 2 = 96). The fact that the RSX is clocked at 550mhz compared to just 500 for the Xenos also gives it an advantage.

Proofreading! You have that backwards, the pixel shaders handle 5 ops per cycle, the vertex shaders 2 ops per cycle (you did it in the right order when you did the math, though). And stop confusing "per cycle" with "per second". If these chips could only do a few ops per second, we'd be seeing pretty awful graphics. :) I'm not trying to give you a hard time, but these things can be confusing enough even with good proofreading. :)

What's also important is to look at what kind of "ops" you're talking about when you say 5 pixel shader ops per cycle on the RSX vs. 2 on Xenos. As I understand from the spec sheets (which may or may not be detailed enough for this kind of analysis), the RSX can do two vector ops per cycle and two scalar ops per cycle, plus fog, whereas Xenos does one vector and one scalar op per cycle. Well, the fog step happens at the very end of the shader -- the output fragment gets blended with the fog color. This doesn't happen every cycle, it just happens "free" at the end of the process. The same thing happens on Xenos, they just don't count it as a 'shader op' on the spec sheet, since it really isn't one. So what I see is 48 pipelines * (1 vector + 1 scalar) vs. 24 pipelines * (2 vector + 2 scalar) + 8 pipelines * (1 vector + 1 scalar). Which gives:

RSX: 56 vector + 56 scalar shader ops per cycle @ 550 MHz
Xenos: 48 vector + 48 scalar shader ops per cycle @ 500 MHz

 

I can see stating that the RSX is the more powerful chip, but let's not overstate it.

 You got me with the proofreading, lol.  When I wrote "per second" I was contemplating adding some calculations dealing with the 500 and 550 mhz clock speeds.  I decided against it, and threw in the line, "The fact that the RSX is clocked at 550mhz compared to just 500 for the Xenos also gives it an advantage."  I guess I just still had the word "second" on the brain.

Also, I was not aware of nVidia counting fog as it's own operation.  If this is true, then yes, the GPUs are far closer than I thought.