NV44a - thoughts and/or predictions?

[Carl B]

I know discussing a low end part like this isn't anything most people on these forums would find exciting, but I'm trying to get a sense of how it might perform in relation to the NV44 from which it is derived from.

Since it will be AGP native, are the chances that it will be able to support a 128-bit bus to 'make up' for the lack of Turbo-Cache? Or will it still be limited to the 64-bit max of the TC? If the former, I could see this part performing on the level of the NV43-derived 6200; otherwise, well....

Thoughts?

 

[DegustatoR]

I vote for 64-bit. Chip's too small for 128 bit bus. They might use somewhat faster memory though.

 

[Carl B]

Well, I hear what you're saying, but from the stats I've been able to find, I'm seeing the NV44 at 110mm^2 on a 110nm process, while the ATI RV350, which supports a 128-bit bus (and incidently is 5 million transistors less than the NV44) is 92mm^2 on a 130nm process.

So though I might believe it could be too much of a hassle to go change the crossbar layout of the NV44 simply for an AGP native revision, I don't think it's the size that would be the constraining factor.

But while we're on the subject, how the hell does ATI make a smaller chip on a larger process?

 

[atomt]

Nvidia numbers does not make sense. From Beyond3D and net, I gather the following:

NV40 = 222M transistors, 0.13um, ~ 287mm^2 (0.13um)
NV41 = 190M transistors, 0.13um
NV43 = 143M transistors, 0.11um ~ 150mm^2
NV44 = 77M transistors, 0.11um ~ 110mm^2

For the same process, die size can be larger if fewer metal layers are employed and layout is not as aggressive. A less aggressive layout could deliver better yields but lower clock speed.

However, I find that NV44 77M highly suspect because even though it have 1 quad less than, the transistors savings from removing 16-bit FP blending & filtering, Z- compression. The PCI-E and Purevideo should be pretty much the same as NV43.