A question about Pixel Pipelines

[Commander_HK47]

Ok, if this has been asked before i apologize for it now.

With saying that, my question about "pixel pipelines" has to do with the now available G80 cores from Nvidia, and the soon to be available R600 Line from AMD.

For the most part i know what they are and what they are used for (pixel texture referencing and more now, because the unified architecture with the pixel, vertex and now geometry shaders)

Knowing that the information on internet in regards to AMD's R600 cores is "hear say" at best until the respective public gets it's grubby little hands on them. I still find it hard to ignore the constant statement that it only has 64 unified shader pipelines as where the GTS and GTX from Nivida have 96/128 respectively.

I even remember reading about how the former ATi at the launch of the X1xxx line of cards said that they felt that a high level of vertex shaders where going to be how the industry would tend to be using shader instructions yet the highest level they have atm is 48(witch is only a multiplication trick on the cards fewer pixel pipelines).

I know pixel pipelines are not the only thing that makes a 3d graphics card go "tick-tock" and thus produce the best looking image at the best frame rate possible(memory type/speed of said memory, memory bus speed, how the card stores particular information:geometry, particle, light, ect).

However in this case it would seem that it would be a major crippling factor if the gape was this large between the two manufactures.

Again if this has been talked about i apologize but i haven't found any topics in regards to this issue.

 

[LuminaryJanitor]

Uh... You said you had a question?

 

[trinibwoy]

All "shaders" are not the same. So just looking at the counts tells you nothing. They are also running at different clock speeds.

 

[defaultluser]

All "shaders" are not the same. So just looking at the counts tells you nothing. They are also running at different clock speeds.

Exactly. The unified pipelines on the G80 processor are significantly simpler than what we're used to, while I expect that the pipelines on the R600 will be closer to the complex processors we are used to.

The 128 pipes of the G80 are clocked at 2.3478 times the ROP clock in order to deliver their incredible performance. And while the shaders have been clocked higher than the ROP units since G70, the clock difference was only about %20. With the 8800 GTX, the clock difference is huge (575 MHz ROP and 1350 MHz shader)!

Best case, the 8800 GTX shows double the performance of the x1950 XTX, but has 2.67 times the "pipelines", clocked twice as fast (2.67 * 2 = 5.33). This would back the assertion that the pipelines (5.33 times "faster" on paper, but only twice as fast in the real-world) are much simpler than what we're used to.

In the end, all you can use to compare them is game performance, because you really have no idea how they will perform simply based on numbers. That, or else youy spend your whole life reverse-engineering the architecture just to estimate performance

 

[vanilla_guerilla]

the secret way that all the experts are using to estimate performance of the r6xx series vs g80 is that they will wait for the goddamn card to come out and read some reviews rather than sitting around biting their toenails over a non-issue. think about something else for a while, if you can.

 

[Commander_HK47]

Yeah sorry, the question was: Will AMD use the past methods to produce there shader pipelines for example the X19xx architecture, or will it be more like Nvidia's approach with regards to there G80 cores and how more the elements have been simplified?

I know that Nvidia used the multiplication measure with ROPs to achieve higher pipeline performance as past methods wouldn't of yielded them that great of a return without a major overhaul in production methods.

I'm not eating my foot in anticipation of the R600 from AMD as i've gotten over than long a go when i first got into personal computing/gaming, I also know in the end that paper performance means didily when comparing it to real time performance because a development team may favor/told to favor aspects of one X company's line of cards of the other .

At this point and time I don't care witch one is going to out beat the other as it won't matter in another 4 - 6 months time, as there will be a new refresh with new features and such for and so on.

I was just trying to gain some clarification on the subject matter as it seemed odd to me that if both of the major graphic card companies where being pushed to a unified architecture that there wouldn't be such a large gap(64/128) [possible meaning that AMD is sticking to the old method of producing independent shader pipelines that acompany the main pipelines] at lest from my surface viewpoint. So i was only trying to get a better understanding of the situation.