All I want to say is that Nvidia needs to get the last few set up drivers up to par for Vista 64 OS and Windows 7 64 OS if they don't want to lose their customer base before FERMI comes out. This also brings to mind why the Inquirer hasn't come out with a story on all those lost NVIDIA cards that fried and systems that won't boot with NVIDIA drivers from 185.xx upward. This has been going on since the end of November 2009.

The PCIe power guideline limit is 300w.

The 5970 slips in at 294w.

By your own words, the weaker GTX360 is 195w. So even doubling the smaller cards power requirement breaks the PCIe compliant guidelines for power draw.

No brainer - if there is a Dual card, it's not going to be following spec guidelines and cant be labelled a PCIe card. Not a good marketing move.

http://techreport.com/articles.x/17986

Astonishing info about Fermi and it real purpose and origin = cut corners or not, please read the following article it all makes sense and make your decisions;

http://www.semiaccurate.com/2010/01/17/nvidia-gf100-takes-280w-and-unmanufacturable/

P.S Truth always reach the surface with the grains of salt ;-D

GeForce GTX 360 TDP 195 and GeForce GTX 380 TDP 225 by current schedule is expected March 2nd - 10th, the specs show it is super, waiting for the benchmarks ;-D.
The GTX 395 - dual GPU Fermi card is not impossible, the question is when will it be launched? Current preliminary info schedules show late April - 10 May 2010. As always all info should be taken with the grain of salt and expect the unsuspected ;-D.

@Bill: It has 512 cores. This is divided into 5 clusters of cores. The 4 clusters are further divided into 4 stream multiprocessors - = 16 SM's.

He did the math again, backwards - 512 cores = 32 cores per stream multiprocessor. (512 cores/16SMs = 32.

There are no 512 SM's, no 16 clusters. Of course since only 448 of the 512 cores, and 28 clusters of cores out of 32 are active according to the NVidia whitepaper, that would now only leave 14 SMs actually working, right?

I hit the wrong button while typing. The 512 cores are divided into 4 clusters. Not 5. Those 4 clusters are divided into 4 SMs each for a grand total of 16 SMs,

Figured I better correct that before some idiotic grammar/math Nazi who can't ass for themself moans about it.

I guess someone didn't slept well last night or is impaired in crtl+c/crtl+v.

"It has four graphics processing clusters and each cluster has four stream multiprocessor (SM) units."

That gives 16 SM. You should try 16 clusters, with 32 SM's in each. If windows calculator isn't flawed, that's 512 SM.
I'm still skeptical about a 512 SM enabled GF100 for the lauch. As even tesla will be cut to 448 SM. Maybe they will launch a 512 SM version, with enough chips to distribute cards to major reviewers. The an actually buyable version, with 448 SM or less.

This GF thing sounds soooo yummy... Hurry up nVidia!

I response to posted by : anon1mat0, 19 January 2010.

If you look at the nVidia whitepaper further you can see that the GF100 has 16 dedicated 'PolyMorph Engines'. Each of these engines has a dedicated tessellation unit that can fetch vertices tessellate and transform them independently from the rest of the pipeline.

Tesellation is done by the same so-called 'cuda' cores and the benchmarks only tested that. In a real world scenario those cores will have other things to do (HD5000 has a dedicated tessellation unit) so the benchmarks may be a tiny bit biased.

We'll see when the card comes out, (plus pricing/availability) it may be faster than ATi (single GPU) but likely harder to find than hen's teeth and significantly more expensive.

Has anyone noticed nVidia have released a whitepaper comparing the GF100 against the Radeon 5870? Claiming from 200% up to 620% better performance from low to high tessellation scenarios.

Dual Fermi?

It has a massive die that will consume a fair whack of power. The GF 100 being touted as better than a 5970 (debatable - wait for real world benchies)would mean consuming at least the same amount of power. A dual Fermi would breal PCIe power regulations and as such could not be sold as PCIe compliant.
There is no talk of a dual fermi (dual card set up but not dual board).
Besids the power draw - costing would be prohibitive.
With the manufacturing at TSMC giving poorer yields than expected, the more complex die of the GF 100 will produce fewer working chips per wafer, increasing costs. So a dual board would be a four figure sum (or sold at a huge loss).
ATI made a fantastically effective card with the 58xx series. The GF 100 if it comes out at all will make a technological statement but i think it will be a gfx card too far and the market for this card isn't there.

The reason no-one is talking about it, is because there is nothing to talk about.

OpenGL 3.2 is the latest standard and is already supported by shipping AMD and Nvidia hardware.

Continuing to support standards is not news.

Why everyone's talking about DX11 only?
Not everyone will use/is using Vista(or Vista SP), you know.

Update

New info from Nvidia shows single GPU Fermi in March around the 2th -10th of March, dual Fermi around late April - May, most likely around 2nd - 10th of May. Schedule's do change and all info should be taken with the grain of salt ;-D.

That would be because Nick only ever rehashes news from other sites. He can't even be bothered to check the details out. Fermi may have 512 cores, but 64 of them will be masked off (whether for yield or heat reasons is unknown).
See http://www.hardware.info/en-UK/news/ymicmJqawpmacZY/nVidia_Fermi_cards_with_448_shaderunits/

Why is the INQ so late with articles? A day or two behind everyone else... there's nothing interesting here anymore.