what is unified shaders?

[PSman]

can somebody here explain to me what is unified shaders?, and why is it so great?. I heard that having unified shaders might make the 360 GPU more powerful than the RSX, is this true?

i apologize if this sound like a dumb question.

 

[TurnDragoZeroV2G]

Being unified only makes it unified, and nothing more or less than that. Everything else is dependant upon how you make them, just like any other shader unit.

In an case, the standard GPU has parts of its logic dedicated to being pixel shaders, and others that only function as vertex shaders. Unified shaders will switch between the two depending upon what work needs to be done. The only bonus, really, is that you don't have one or the other, ideally, sitting idle waiting for the other group to finish their work. So, higher efficiency is possible, but that's pretty much it.

 

[PSman]

so having unified shaders doesn't make the 360 GPU more powerful?. It just make it more efficient?

 

[Razor1]

yes thats the gist of it. Well it might be more effecient, it all depends on how effecient its scheduler is.

 

[Mintmaster]

so having unified shaders doesn't make the 360 GPU more powerful?. It just make it more efficient?

Well, if you're more efficient, then you're more powerful given equal resources, right?

The 360 GPU (Xenos) will have certain advantages over RSX due to it's unified architecture. If you have complicated vertex shaders (like character skinning, etc) then Xenos will have much higher polygon throughput than RSX (up to 6x theoretically). If you have texturing in the vertex shader, Xenos will be much faster (order of magnitude). Both of these statements, though, assume your vertex load is limiting performance, and that RSX will be similar to G70.

That's pretty much covers the major advantages of a unified architecture. It can blast through computationally intensive vertices, especially those without pixels (i.e. on the hidden side of an object or offscreen), so that the pixel shader stays fed.

 

[PSman]

wow!, thanks everyone I've learn a lot

I got another question, hope you all don't mind. Can you guys tell me more about the RSX?, I heard that it is base on the 7800gtx, but it will be a lot more powerful than the 7800gtx(don't know if this is true or not, though). And does the RSX have any kind of advantage over the 360 gpu?

thanks a lot

 

[Mintmaster]

We don't know that RSX is similar to G70, but we're guessing it because the marketing numbers from NVidia, like pipelines, fillrate, and shaderrate, all match with G70.

RSX's advantages are a higher fillrate with only 2xAA or no AA, and faster filtered texture processing. These advantages may not be realized often, though, because they all require bandwidth to achieve, and RSX has only 41% of the bandwidth to the GDDR3 memory as the equally clocked core of the 7800GTX 512MB. People point out the connection to the XDR memory, but if you're using even a fraction of Cell's potential, that will not be available.

I'd say for traditional graphics stuff, RSX will come out on top if developers can compress textures and make do with 32-bit framebuffers. For newer rendering techniques involving HDR, branching shaders, lots of math, and vertex texturing, I'll take Xenos. However, these tentative guesses aren't blowout wins for either architecture. IMO, it will all come down to developer and artistic talent, and the hardware differences will just be a wash.

 

[supervegeta]

can somebody here explain to me what is unified shaders?, and why is it so great?. I heard that having unified shaders might make the 360 GPU more powerful than the RSX, is this true?

i apologize if this sound like a dumb question.

That's not a dumb question.

Unified shaders means the gpu don't have separate vertex and pixel shadres but all the
48 unified alu can do either vertex or pixel operation.

This will make Xenos more efficent but not more powerful than the RSX.

 

[Mat3]

So how would the performance of a single, traditional vertex shader compare to a single unified shader, just for the sole purpose of vertex processing? Same question on the pixel side...

Is there a difference? Or, what, if any, is the downside to a unified shader architecture?

 

[Razor1]

As said before there really shouldn't be a down side with performance of unified vs. individual units, other then scheduler overhead. The only disadvantage for unified would possible be more control silicon neccessary for the control over the units (again scheduling). But performance will see a boost if some any of the shader units gets bottlenecked for certain operations in unified.

 

[czekon]

I was wondering about one thing, if Xenos is USA gpu can it be possible to gain extra pixel shader power by delegating some vertex work to the CPU??. I know, that some devs are using PS to generate trees for exemple, like speedtree. I dont know how big offloading it will be, and how many ALU you get free to do pixel shader work. Is it even worth to try?, or getting few ALU will not change anything, will only cost you extra work and waste of CPU resources that you can spare on something else : like physics.