How Many Fragment Pipelines Does G70 Really Have?

The logical conclusion would be to kill the 6800U, leave the 6800GT and the vanilla 6800 in place, create a 7800 vanilla to fill in the spot the 6800U left behind. If anything, they could probably add a few clocks on top of what the 6800U did, and have a little bit of a performance difference compared to the 6800GT (stock).
A 16 pipe 7800 @ ~450 would make a decent replacement for the 6800U at the same price or slightly lower. Of course the logical time to do this would be during the ATI R520 launch, alongside a new high end part. The 7800GTX owners would have had a full quarter or so with their card as top dog, although some will feel cheated anyways.

new fall lineup:

6800 NU - 199 (12 pipes)
6800 GT - 299 (16 pipes / 350Mhz)
7800 NU - 399 (16 pipes / 450Mhz)
7800 GTX - 499 (24 pipes / 430-450Mhz)
???? U - 599 (?? pipes / ? Mhz)


Where NU = Non Ultra, or vanilla.

 

Nv seemed to make a big point of the fact that they have the same # of potential dies per wafer at this larger-than-NV40 size. It seems to me the flip side of this is that if it was the same size as NV40 they wouldn't have been able to have more dies-per-wafer.

Given the above, is it possible they intentionally made the transistors on the chip less dense than they could have, for heat-dissipation and signal-interference reasons? Or does using "standard cells" (rather than custom ones) mean this just isn't possible?

 

The new leaked 80.40 drivers make reference to a G72. The supposed "ultra"? There is also a C51....

Quote from inf file

NVIDIA_G70.DEV_0090.1 = "NVIDIA G70"
NVIDIA_G70.DEV_0091.1 = "NVIDIA GeForce 7800 GTX"
NVIDIA_G70.DEV_0092.1 = "NVIDIA G70 "
NVIDIA_G70.DEV_0093.1 = "NVIDIA G70 "
NVIDIA_G70.DEV_0094.1 = "NVIDIA G70 "
NVIDIA_G70.DEV_009D.1 = "NVIDIA Quadro FX 4500"
NVIDIA_G70.DEV_009E.1 = "NVIDIA G70GL "
NVIDIA_G72.DEV_01D0.1 = "NVIDIA G72"
NVIDIA_G72.DEV_01D1.1 = "NVIDIA G72 "
NVIDIA_G72.DEV_01D2.1 = "NVIDIA G72 "
NVIDIA_G72.DEV_01D3.1 = "NVIDIA G72 "
NVIDIA_G72.DEV_01D5.1 = "NVIDIA G72 "
NVIDIA_G72.DEV_01DE.1 = "NVIDIA G72GL "
NVIDIA_G72.DEV_01DF.1 = "NVIDIA G72 "
NVIDIA_NV48.DEV_0210.1 = "NVIDIA NV48"
NVIDIA_NV48.DEV_0211.1 = "NVIDIA GeForce 6800 "
NVIDIA_NV48.DEV_0212.1 = "NVIDIA GeForce 6800 LE "
NVIDIA_NV48.DEV_0215.1 = "NVIDIA GeForce 6800 GT "
NVIDIA_C51.DEV_0240.1 = "NVIDIA C51GL"
NVIDIA_C51.DEV_0241.1 = "NVIDIA C51"
NVIDIA_C51.DEV_0242.1 = "NVIDIA C51 "
NVIDIA_C51.DEV_0243.1 = "NVIDIA C51 "
NVIDIA_C51.DEV_0244.1 = "NVIDIA C51 "
NVIDIA_C51.DEV_0245.1 = "NVIDIA C51 "
NVIDIA_C51.DEV_0246.1 = "NVIDIA C51 "
NVIDIA_C51.DEV_0247.1 = "NVIDIA C51 "
NVIDIA_C51.DEV_0248.1 = "NVIDIA C51 "
NVIDIA_C51.DEV_0249.1 = "NVIDIA C51 "
NVIDIA_C51.DEV_024A.1 = "NVIDIA C51 "
NVIDIA_C51.DEV_024B.1 = "NVIDIA C51 "
NVIDIA_C51.DEV_024C.1 = "NVIDIA C51 "
NVIDIA_C51.DEV_024D.1 = "NVIDIA C51 "
NVIDIA_C51.DEV_024E.1 = "NVIDIA C51 "
NVIDIA_C51.DEV_024F.1 = "NVIDIA C51 "

End quote

Obviously a revision is clearly planned (nothing new). But its interesting...

 

C51 is the next nForce motherboard chipset, I believe.

 

Presumably integrated graphics?

Jawed

 

According to Anand, yes:

C51 mention

I'd assume this will contain support for SM3.0 and hopefully the video assist engine too (hopefully, a fully-functional engine this time, too ).

 

Thanks for the info about the c51. An intersting read.

 

Yeah, but when are you going to tell everyone that it's half a G70, with TurboCache support?

 

After you

 

Haha tEd was really spot on

 

What's that with 90 nm? And where's the quote from? :?

 

G72 is for the midrange card

 

Link from DB

 

G70 Roadmap

nVidia G70 90nm GPU is expected to be released in the Fall. This GPU will be based on the G70 core, but will feature enhanced clock speeds due to the reduction in core size resulting from the move to 90nm.

nVidia G7x Mainstream GPU is expected to be released in the Fall. This GPU is expected to be a cut down version of the G70 90nm core.

nVidia G7x Budget GPU is expected to be released in the Fall. This GPU is expected to be a significantly cut down version of the G70 90nm core.

 

EntombeD: link? Or at least credit.

Hedging their bets with a 110nm G70 seems like an eminently smart move. Perhaps costlier in one respect, but look at what ATI's going through getting R520 up to snuff. They'd have been better off from a marketing and probably financial perspective had they a 130/110nm R520 to plug the gap until they got 90nm working well.

 

Hasn't ATI said that it needs 90nm to go to SM3?

Maybe that "restriction" only applies at the high end - since mid-range and low-end SM3 parts would prolly have much smaller transistor counts.

Maybe ATI's counting on 90nm for margins, too - getting a margin increase even at the low-end from 90nm.

Jawed

 

VR Zone is reporting 7800GT at 20 pipes/400mhz. If true, that would seem to put a thumb on the scale against the "hidden quads" theory. Agreed?

 

Interesting turn of phrase Geo, I take it you're suggesting that a 20-pipe G70 does not lend any credence to the "hidden quads" theory?

Would a 5-quad G70 setup be likely to have efficiency problems in the same way that the X800 Pro had for a few months after its introduction until these were improved by (IIRC) crossbar optimisations and drivers updates? I don't recall any talk of the vanilla 6800 (3-quads) having similar inefficiencies as the X800 Pro initially, so perhaps nVIDIA were able to resolve the issue before release and/or better hide any additional latency within their GPU architecture?

Cheers,


BrynS