Yes but exactly what is a pipeline? My card has 20 pipes but they are arranged in groups of 4 and feed only 16 ROPs so there are only 16 complete paths thru the card. Many are calling the 1900 a 48 pipeline card because it has 48 pixel processors. Is this wrong?
It depends on the circumstance. Simplistically you would think more pipelines would be faster since more work can be done in parallel, but that doesn't always pay off right now in terms of efficiency. As it stands, the architecture isn't so much limited by pixel pipelines as by the render backends and texture units. The hope is to offload tasks from the TMUs to shader programs which not only frees up the TMUs, but also allows for more advanced graphics. However, doing so doesn't just require additional pipelines, those pipelines need to have the processing power to handle additional complex shader programs which is why there is a 3:1 ratio of pixel shaders to pipelines.
It isn't just ATI, nVidia is moving toward this direction as well, but they don't agree on the ratio. ATI wants a 3:1 ratio, while I believe nVidia wants a 2:1 ratio. It gets kind of confusing though, because nVidia is keeping the pixel shaders and TMUs matched at 32 each, while it only has 16 ROPs. ATI of course has 48 pixel shaders, and TMUs and ROPs matched at 16.
Marketing scams? What's the scam? It's just changing technology.
A pixel pipeline can process one pixel per clock, I believe. So if you have 8 pipelines you can process 8 pixels per clock, 16 pipelines and you can process 16 pixels per clock, etc...
But the old way, you would have to re-process a pixel if it required more operations than it could handle in that clock.
So even if you had 16 pipelines, it might take two or three clocks to finish applying shaders to them.
From what I understand, now that the shaders are indipendant of the pipelines, you can process a pixel in one clock with a whole lot of shader operations, and probably won't have to reprocess that pixel a second time even in complex games. Which would contribute to efficiency.
It has 16 pipes.. with three pixel shaders at the end of each Pipe.
the pixel shaders take more time, thats why ati decided to put 3 at the end instead of 1.
16x3 = 48 pixel shader operations per clock cycle.
its still a 16 pipe card, it just has better SHADER performance.
the 48 pipe statement is a marketing gimmick used to confuse people who
think more pipes = greater performance.
The performance gain isn't exactly relative... b/c they've confused pixel shaders with pipes. the total processing power is still limited by the 16 pipelines, althogh now you can process the pixels at the end more effieciently. (3 pixel shader operations instead of 1, which should really help with unlimited pixel shader operations)
A 48 pipe radeon (esp w/ 3:1 ps) would slaughter any 2 nvidia cards in sli you can think of.
I agree, although you can get up to 48 pixel shaders does not mean you have 48 pipelines. Therefore, its false advertising no matter what the excuse. If you think people should know that they can get 48 pixel shaders then tell them that not pipelines.
