nv47 to have 24 pipes ???

[MUYA]

having great architecture is one and having to use a mediocre manufacturing process is another thing. There is also the question of low ns memory module availabilty. I really dun think that a 6 quad processing GPU is what the nv47 is....due to the reasons mentioned above

[Graphicmaniac]

sometimes ago i have read an article where somebody of Nvidia said thatthe future of Gpus is go over the Ghz.

Here we are still talking of 400..500.or at max a 600 mhz gpu

does nvidia have some secret plan to use a tecnology that goes over ghz barrier or maybe they are choosing to follow AMD into this new era of multicore tecnology?

they have done sli, next steep could be multicore.. or not?

[nutball]

sometimes ago i have read an article where somebody of Nvidia said thatthe future of Gpus is go over the Ghz.

With GPUs the companies have a choice to go fast (up the clock), or go wide (add more pipelines). Doing both in a single chip is probably impossible at the moment, at least both both in extremes.

The CPU guys don't really have the option of going wider (adding more functional units) because they're already at or near the point of diminishing returns in intstruction-level parallelism. So they go fast, and pump up the clock-speeds.

For GPUs there's so much untapped parallelism, it's still profitable to add functional units. It seems that for now that's what both NVIDIA and ATI are content to do, to take the option of adding more pipelines and making only moderate clock-speed increases.

[retsam]

wow nutball that really the nail right on the head

[harl]

does nvidia have some secret plan to use a tecnology that goes over ghz barrier or maybe they are choosing to follow AMD into this new era of multicore tecnology?

they have done sli, next steep could be multicore.. or not?

Multicore it's a mosense in a GPU, because they can improve paralelism
with more pipes
A 2 core GPU with 32 pipes will be the same than a 32 pipes single core
GPU and maybe mor complex.

But maybe they should try embbede RAM or somthing becasue ram
it's turning into a bottleneck,

[Nutty]

Yeah embed the frame-buffer into the core is a good idea. The gamecube already did that. You improve fillrate immsensly with that. Its also good if you want to add general frame-buffer readback in the pixel shader. That bottlenecks the core at the present moment, which is why you cant do it.

The problem is, you need a ton of memory for PC cards because we all expect 1600x1200 with 4xAA on nowadays, and that chews up a massive amount of RAM.