GeForce FX rumored to be 4x2

[silence]

looks like guys in that magazine did their job very good.it was 11/2002 or 12/2002 issue,just after paper launch.
they were also worried about "loss less compression" pointing out that with 128 bit bus FX might not be as fast as nvidia wants it.......i'll try to dig out that article and see how much they guessed.

[Darth Rancid]

The FX being 4x2 explains why Anand didn't gain any significant performance by overclocking its memory interface I guess...
It isn't as bottlenecked by bandwidth as we thought.. its lacking good old fashioned pixel fillrate!

Now, I admit to having lost the grasp on GPUs since the appearance of the GTS and its arcane "shaders"... I like to think in terms of Pixel and Texel fillrates along with triangle throughput and memory bandwidth.. so my knowledge on this matter is a bit outdated.
But here's my possibly faulty conclusion:

* The FX performs "as it should", the hopes that it would gain massively with the next driver set is effectively dashed.

* For the NV35 to have 8 pipes would require either an insane transistor count, even more so than the NV30. Or that the NV30 is actually a 8x2 but 4 pipelines are broken. This would explain the insane transistor count of the FX.

Well.. thats my thoughts at this time...

[StealthHawk]

Originally posted by Darth Rancid
The FX being 4x2 explains why Anand didn't gain any significant performance by overclocking its memory interface I guess...
It isn't as bottlenecked by bandwidth as we thought.. its lacking good old fashioned pixel fillrate!

except that it is bandwidth limited where it counts, high resolution 4x FSAA/AF.

gfFX doesn't have many problems beating r9700 in AF or 2x FSAA, or in raw performance, only when 4x FSAA/AF are thrown into the mix.

[The_KELRaTH]

I would update that to say the GFFX, in 12bit Int mode, doesn't have many problems just beating r9700 thats in 24bit FP mode.

[Bigus Dickus]

Looks like this topic should be moved from the "rumor mill" back into the main nVidia cards forum. :)

For all practical purposes on any existing game, the GFFX is a 4x2 architecture. For some future games it will act like a 4x2 sometimes and an 8x1 sometimes.

Did nVidia do screw up anything else, or do we know the whole ball of **** now?