Let's look into the future..

[TecHNooB]

This pertains to vid cards of course..

I don't know about some of you guys, but I love the idea of having more pipes than higher clock speeds. To me, higher clock speeds is more stress on the video card. More pipes is like having a bigger highway. Right now, NVidia is the one concentrating on adding pipes. ATI is the one cranking up clock speeds. Their answer to NVidia's added pipes? Fat pipes. Right now, fat pipes don't increase performance by much because the pipe numbers are small. But what happens when we have 128 pipe cards? The fat pipes will truly shine when the pipe numbers get that high. So 10 years down the line, ATI's fat pipes will stomp NVidia's normal pipes. Agreed?

Flame suit on.

 

[shabby]

Yes agreed... fatty!

 

[overst33r]

Originally posted by: TecHNooB
This pertains to vid cards of course..

I don't know about some of you guys, but I love the idea of having more pipes than higher clock speeds. To me, higher clock speeds is more stress on the video card. More pipes is like having a bigger highway. Right now, NVidia is the one concentrating on adding pipes. ATI is the one cranking up clock speeds. Their answer to NVidia's added pipes? Fat pipes. Right now, fat pipes don't increase performance by much because the pipe numbers are small. But what happens when we have 128 pipe cards? The fat pipes will truly shine when the pipe numbers get that high. So 10 years down the line, ATI's fat pipes will stomp NVidia's normal pipes. Agreed?

Flame suit on.

who's to say that nvidia wont have more pipes to balance the loss in fatness...

 

[Pete]

What's a fat pipe? The G70/7800 seems to have a "fatter" pipe in that it's capable of more calculations per clock than the R520/X1800 thanks to its second ALU supporting both MADD [sic] and MUL (while R520's mini-ALU supports just ADD), AFAIK.

Both are aiming for the highest profits, and that generally means that a smaller die size will yield more cores per wafer, so smaller but higher-clocked may be more desirable from a manufacturing POV. Of course, both the 9700P and now the 7800GT bucked that trend (as did the 6800GT, but to a lesser extent, as it was on the same process as the X800P--but still lower-clocked).

Despite the G70's relatively fatter pipes, both companies are trending toward more and simpler ones in some ways. NV30 -> NV40 added pipes but of suppsedly lower complexity. R500/Xenos isn't as easy to compare, but it offers 48 identical ALUs, whereas R520 (like R420 and R300) has 16 shader units that consist of a main and a mini-ALU. If ATI is moving toward unified shaders, I'd expect to see something like R500: a few large shader "pipes" packed with lots of single, "general purpose" ALUs.

Or are you talking about R580 having far pipes compared to R520? This seems to be heading in R500's unified shader direction, with more ALUs available per higher-level unit (which I guess is still a quad in R520/580's separate pixel and vertex shader cores).

Warning: I may be talking out of my ass toward the end, but I'm reasonably sure about most of this post. And, obviously, all these GPUs could use more cowbell units.