Ati vs. Nvidia Unified Shaders

[ekoh]

Why the huge gap between ATI and NVIDIA on unified shader? Looking at two comparable Cards the GTS 250 Has like 126 while the 4870 has 800. What am i missing here?

 

[Bobnova]

Despite calling them the same thing, they are very different beasts.
The nvidia flavors run at a much higher clock speed and do more per clock, while the ati flavor simply has MORE.

The end result is about the same for most things graphical, but some GPGPU things (like [email protected]) vastly favor one over the other (in folding's case, nvidia works much better at the moment).

 

[Vurtexx]

AMD has a number of units, which if I understand it right partly have to rely on software to really show their potential, because with the wrong software some of that shader power goes to a sort of waste because of the design.

NVIDIA's approach is a little different... they have less of what a person might call shader units and they (or at least most of the modern cards) have their own clock speed which is ran faster than the core.

When you compare 128 NVIDIA shaders to 800 AMD shaders, I think when you figure it in that way each NVIDIA shader is more powerful... there's both a higher clock speed, and probably a higher efficiency, to each NVIDIA shader considered... though it's possible the 800 shaders can still deliver more power than the 128 NVIDIA shaders (and probably would in the right shader-intensive programs), since there's still a lot of potential for brute force from the 800 shaders.

 

[Firestrider]

If you want a more in depth analysis of each architecture look here:

RV770:
http://www.rage3d.com/reviews/video/atirv770/architecture/

GT200:
http://www.realworldtech.com/page.cfm?ArticleID=RWT090808195242
http://www.beyond3d.com/content/reviews/51