Question about NV2A (comparing to C1)

[AlNom]

I just want to get an idea of the jump between the two console graphics chips:

1)How would one compare the shader power between the two? (in terms of ALU operations that can be done per cycle)
2) What is the NV2A equivalent to a "shader pipe" And "how many" are there in NV2A?
3) Does performance scale as per-pixel-per-shader operation?

Thanks

 

[BRiT]

For a rough idea, compare a GF3-TI/GF4 to a 7800GTX.

 

[akira888]

NV2A, like the GF4Ti has two vertex shader units. Also, according to the various NV2X programming guides Nvidia distributed it would seem that each of the four pixel pipelines had two register combiners, with the capability of coissung a vec3 + scalar color blends per cycle on each combiner. Fill rate falls by 50% with 3-4 ops, 67% with 5-6, 75% with 7-8 (the maximum).

 

[AlNom]

ooo... that's interesting. I also didn't know they released programming docs for the 2A. Thanks

Are the docs on their developer website?

 

[akira888]

Well, not docs for the NV2a, but for the NV20/25.

Here's a basic overview of the NV2x architecture. Section 28 elucidates the relationship between the number of combine ops and the effect on theoretical fill rate.

http://developer.nvidia.com/object/geforce3_faq.html

 

[Heinrich4]

I read certain time that NV2A reached about 28 flops for cycle(pipe vertex = 2 x 8flops + 1 Pixel Shader = 12 Flops, and if not i wrong with 2 ALUS to per pipe and 2 z-stencil buffers,capable de tessalation ) that it must give 6.5GFlops (233MHz) or 1/44 of the power of the C1/Xenoss (240Gflops to shader pipes + 48 GFlops fixed functions of eDRAM).

 

[AlNom]

Thank you for the info.

 

[Powderkeg]

NV2A is almost the same thing as the GeForce4 Ti 4200. Same clock speed, same memory bandwidth (Of course the Xbox is shared memory) same number of shaders, there is not a lot of difference between them.

 

[pjbliverpool]

NV2A is almost the same thing as the GeForce4 Ti 4200. Same clock speed, same memory bandwidth (Of course the Xbox is shared memory) same number of shaders, there is not a lot of difference between them.

Actually it was clocked slower and had less memory bandwidth. It also lacked most of the efficiency upgrades that NV25 gained over NV20. I think it lacked PS1.2/1.3 aswell.