How about some speculations about AMD R680_65nm

In all seriousness, I agree with that...accept of course the clockspeeds/mem. It seems ATi is targeting somewhere in the 700-800 range for the R600, so one would assume it's close to the 1ghz mark.

The questions that are burning though are:

1. Will it be a straight shrink to 65nm, ala to get the size down (to < 300mm2), or will they change the architecture by adding more shaders/tmus/whatever, or by redefining those shaders (perhaps something along the lines of 64->96, or vec4->vec5), or will perhaps both be possible and done?

2. Are those old rumors of R600 running at 1.2v to the core true? Because if they are, and R600 runs hot even at that spec (because of it's size, setup, clocks, whatever, at least on A12) it would seem that that spec leaves a lot of room for boosting come revisions. We know ATi's played that game in the past with leverging parts based on stock core voltage (XL/XT). Havn't high-end GPUs in the past been in the 1.4-1.5v range with mid-range at 1.2v? If R600XT/X comes out at 1.2v, perhaps R680 could be a straight shrink to 65nm with 1.4v stock or something, with the smaller area allowing the heat to be reasonable for such voltage, resulting in massive clocks.

3. What spec of GDDR4 will be available in Q3/Q4 of this year? 3.2ghz? 3.6ghz? We've heard both will be available, and that yields are good, yet adoption seems to be slow and samsung is slow to increase bins and Hynix is yet to even come out with something.

 

Renaming the products had nothing to do with competition or architecture, going from NV50 to G80 doesn't give you a 50% speedboost.
If AMD will anounce it as a FX2800, it won't run faster than a r600

 

No, I'm sorry, you fell victim to the marketing machine.
Read the G70 technology preview (http://www.beyond3d.com/previews/nvidia/g70/)

The name G70 was used as a clean slate for nV's product numbering scheme NV47 is G70 and no theory will change that. NV50 is G80.

No

NV40 was the GeForce 6 line, not NV45, NV45 was native PCI Express instead of AGP, that's all. G70 really isn't all that much different from NV40 hence a small step from NV40 to NV47.

 

Small correction, NV45 was not native PCI-Express.
I have one and can confirm it's still the good ol'NV40 AGP8x, bridged to PCIe with the BR02 chip on-package.
The fact it shared the package with the GPU was what gave Nvidia the opportunity to call it "NV45", not to mention reusing most of the PCB components from the AGP line.

But the outcome was acceptable, since the name is still "6800 GT/6800 Ultra", and the translation chip is no perceived bottleneck in the apps i've been using it.

 

Err, didn't the pcie variant not have the PureVideo problem? Wouldn't that indicate it was a different chip?

 

It's somewhere right in between R300-R350 and R520-R580

Performance Doubles? hardly at all. maybe some benchmarks will be better, but the only thing that is close to doubling is the shader and texture performance, simply because G70 has 24 units and NV40 has 16.
The pixel fill rate etc. hardly increased at all. Memory bandwidth increased 10%
Really, check the reviews done here and then come back, G70 was no big step at all.
http://www.beyond3d.com/previews/nvidia/g70/index.php?p=07

I still think it's R650 btw.

 

You mean the borked Windows Media Acceleration? that was NV40. but I have lost all track on which models did have the problem and which not. seems like all spins in NV45 were working and NV40 were not. Thus making it a AGP only problem (what a clever way to get you to buy a PCIe board eh?)

 

Yeah, that.

 

Different revision for sure (even if only due to the extra chip on package).
Purevideo was announced before the PCIe version came out, so it's easy to infer that they did one.

But that does not invalidate the fact that it's still NV40, or else they would use a truly native chip, instead of the bridge.

 

you might be thinking of the non GT 6800, which was NV40 on AGP, but NV41 and NV42 on PCIe.

there's also a NV48 which is still a NV40, right? what is it for? (I remember reading it was for the 6800 ultra 512MB but that seems stupid. btw did anyone really buy that 700 euros, late and useless card.)

 

there is absolutely no reason to have 2 gigs of memory on a gfx card at this time or in the next 2 years. period. 1 gig maybe in a year. very few games need more then 256 to work well. 512 for extreme high resolutions which most of us don't play with probably less then 10% do. and you can make almost any game work well with 512. 2 gigs of textures is stupid and unneeded and unheard of...

game designers build for what the average user has which now and in the next year will be 256.

 

doom 3 ultra quality *cough*

If game designers build for what the average users have, everything would be super smooth on my 64 meg shared memory i915G.

Software sells hardware, NOT the other way around.

 

It sells a small amount of hardware to a small number of people.

Code:

[B]VRAM  (945512 Users)[/B]
32 MB     21,868     2.31 %
64 MB     88,783     9.39 %
96 MB      4,959     0.52 %
128 MB   321,428    34.00 %
256 MB   396,133    41.90 %
512 MB    75,612     8.00 %
Other     36,729     3.88 %

From here.

Doom 3 ultra-quality didn't represent the baseline hardware configuration for which the game was designed. If it did, they wouldn't have sold many copies. D3UQ is added bonus content for the small number of people who have cards which can cope with it, perhaps it makes good benchmark fodder too which keeps the game in the public eye for years to come.

 

Like all the people looking for Vista compliant PC's?

The steam survey is a pile of steaming dog droppings. Are you sure the Geforce 8800's outsells all i915GM or i945GM chips? (hint, my company owns 2500 intel graphics based laptops)

Code:

[i]DX10 Grahpics card vs. something you wouldn't even advise your mother to buy[/i]
NVIDIA GeForce 8800	[color=green]5,964[/color]
Mobile Intel 915GM/GMS, 910GML	[color=blue]5,323[/color]
Mobile Intel 945GM Express Chipset Family	[color=blue]4,984[/color]

People buy performance cards nowadays..
Two or three generations ago 512MB cards allready sold like hotcakes (512MB 9600's), it's people still on geforce 4's that mess up the statistics. They can't run DX9 games but they can run a lot of Steam's catalogue and make others think we actually like to see our games optimized for 64 and 128mb vram

 

I do agree that software sells hardware but I'm really trying to understand why we should care about the configurations of corporate workstations I'm quite sure "average user" means average gamer and not average computer user.