Since no one on nVnews knows...

This may be a valid hint concerning NV35's fragment shader pipeline count:
I think that the NV35 you describe could be a good chip but that a NV35 with 4 pipes could also be good. Of course, the R3x0 would remain more powerful in FP. I personnaly think that NVIDIA won't show FP power as the main capacity of the NV35. They'll still ask developers to use FX12 because it's necessary for the GeForce FX line and they need that games run well on every GeForce FX not just on the NV35.

Also, the fact that the change between NV30>NV35 mirrors NV20>NV25, indicates something. NV20 maintained the same number of pixel pipes, but recieved other efficiency enhancements, such as a new memory controller, extra PS flexibility, and another VS.
 
I know I'm insistant, but anyway... 8)

The biggest problem with the high INT performance is that DX9 *doesn't let you use Integer in PS2.0.+*
That means the only way to use integer is using nVidia's proprietary OpenGL extensions - or use Cg & OpenGL. And D3D developers are not gonna switch just to please them.

What nVidia is doing, though, is pretty much coding things in a language which I suspect to be very near of the real architecture, and then give such a shader to developers to integrate them to their games. That means they're bypassing the API, and an use FX12 in DX9.
Stalker is an excellent example of such a game. But I'd be surprised if nVidia could manage to do that with every developer... Quite a few, perhaps...

Heck, I suppose you got more info than me about it, so I won't insist further. But I'd still be highly surprised if the NV30 didn't have 8 true pipelines.


Uttar
 
Uttar said:
I know I'm insistant, but anyway... 8)

The biggest problem with the high INT performance is that DX9 *doesn't let you use Integer in PS2.0.+*
That means the only way to use integer is using nVidia's proprietary OpenGL extensions - or use Cg & OpenGL. And D3D developers are not gonna switch just to please them.

What nVidia is doing, though, is pretty much coding things in a language which I suspect to be very near of the real architecture, and then give such a shader to developers to integrate them to their games. That means they're bypassing the API, and an use FX12 in DX9.
Stalker is an excellent example of such a game. But I'd be surprised if nVidia could manage to do that with every developer... Quite a few, perhaps...

Heck, I suppose you got more info than me about it, so I won't insist further. But I'd still be highly surprised if the NV30 didn't have 8 true pipelines.


Uttar

I was thinking exactly like you Uttar. And I couldn't think of a so limited FP power in the NV30.

But... At Cebit, NVIDIA gives me "The Cg Tutorial" book, a nice present. (A great book to begin!). I read it after the Cebit. Some small details caught my attention. It's why I managed to find some answers about GeFFX's pipelines. Actually, NVIDIA says :

"The CineFX architecture also supports a special high-performance continuous data type called fixed for fragment programs. The fixed data type has a [-2, +2[ range (meaning, ranging from negative 2 to not quite positive 2) for fp30 profile. In other profiles, the fixed data type is synonymous with the smallest continuous data type available. Although the Cg compiler (cgc) and runtime support the fixed data type (and vector versions such as fixed3 and fixed4), Microsoft's HLSL compiler (fxc) does not."

fp30 is NV_fragment_program.
ps_2 is DX9 fragment.

And in the Profiles table, NVIDIA says :

"
fp 30 : fixed : Fixed-point with [-2,2[ range
ps_2_0 and ps_2_x : fixed : Depends on compiler settings
"

I found that was… strange. Microsoft doesn't talk about fixed in DX9 pixel shading. But NVIDIA says that Cg can compile DX9 shader with fixed ! I wanted to find some details about that and it's why I found the 1FP unit and 2FX units story. So with Cg, you can use fixed in DX9. It wouldn't surprise me if NVIDIA and Microsoft were having many "nice talks" about that ;)
 
Back
Top