Earlier in December we reported that NVIDIA had more than likely lost the big high-performance computing win with Oak Ridge Labs, and the lack of any kind of response pretty much confirms that. 

UPDATE: NVIDIA Fermi continues to have problems according to multiple sources - Graphics Cards  1
The Fermi Architecture – 512 cores no more?

Well, as the holiday approached and then passed us by, the news does not appear to be getting any better for NVIDIA.  This story on Semiaccurate claims that NVIDIA has *ahem* lowered the number of shader processors on the architecture from 512 to 448 – a drop of nearly 15% of the compute power of the GPU.  The SA article quotes from a Tesla product PDF that mentions the shader count and shader speed (1.4 GHz) along with a board power listing of 225 watts.  Consider that the last 10% of any GPU/CPUs performance costs well more than 10% of the total power, had NVIDIA wanted to keep the 64 shaders it is apparently dropping on these first products it would easily be 260+ watt card!

We are also hearing news that the GPU is being delayed once again into March of 2010 – this is a quite a bit later than the initial November 2009 release planned during the summer.  Another potential time frame was even said to be at next month’s Consumer Electronics Show but that also appears to be out.    

Here is what the Digitimes story has to say:

Nvidia originally scheduled to launch Fermi in November 2009, but was delayed until CES in January 2010 due to defects, according to market rumors. However, the company recently notified graphics card makers that the official launch will now be in March 2010, the sources noted.

Nvidia plans to launch a 40nm GDDR5 memory-based Fermi-GF100 GPU in March, and will launch a GF104 version in the second quarter to target the high-end market with its GeForce GTX295/285/275/260, the sources pointed out.

For performance level markets, Nvidia will allow its GeForce GTS250, GT240/220 and 9800GT/9500GT defend against AMD’s Radeon HD 5770/5750, 4870/4850 and 4670/4650.

For the mainstream market, Nvidia will mainly push its GeForce 210.

If we combine these two stories you can see how bad the situation MIGHT be for NVIDIA: not only is Fermi going to be late but it could be notably underpowered compared to where the company’s initial estimates placed it.  Not only that, but the above rumor even puts the GF104 part, for the higher end gaming markets, even later in Q2 of 2010 while leaving the GTS 250 and 9800-series in to battle the AMD 5000/4000 series.  Keep in mind that the GTS 250 has essentially been the same GPU that was released December of 2006!!

If you were tired of rebranding already – tough luck it would seem for the first half of 2010.

One company that is loving all of this – AMD and the ATI team responsible for the Radeon 4000 and 5000 series cards.  They have obviously made the better choice in terms of creating a solid and balanced GPU roadmap for today and the future by focusing on smaller, scalable GPUs rather than immense monolithic GPUs like NVIDIA. 

UPDATE: NVIDIA Fermi continues to have problems according to multiple sources - Graphics Cards  2
How long will it be until NVIDIA can compete with the Radeon HD 5970?

You can be sure we’ll be asking the tough questions at CES in just a week or so to get some more answers; if not from NVIDIA then from anyone that will talk. 

UPDATE 12/30/09

Last night I had a conversation with someone from NVIDIA that wanted to dispute some of the claims in our story as well as put in their view on some of the issues at work here.  First, NVIDIA wanted to say that there has been “no change” in their relationship with Oak Ridge Labs and the use of Fermi in the organizations upcoming supercomputing product.  NVIDIA pointed me to this story that claims to have spoken with representatives from Oak Ridge:

A little later we got on the phone with someone in Computing and Computational Sciences at Oak Ridge National Laboratory and they stated that the SemiAccurate article was inaccurate and had no further comment. Legit Reviews also contacted Andrew Humber over at NVIDIA who informed us that the original press release that was issued in September is still valid and that nothing has changed.

While that isn’t the most forthcoming quote from the computing legends, I guess it’s something.  What we don’t know is that if any of the behind-the-scenes details have changed since the initial announcement back in the fall.  Obviously neither company would really come out and say if something totally off the wall had been discovered and instead they would work through the issues behind the scenes.  That leaves me with a he-said he-said debate but hey, that’s all we can report on.

As for the part of our story about the slightly less powerful Fermi GPU that was announced and discussed at this past week’s supercomputing convention, NVIDIA’s reps wanted to clearly point out that there is NOT a direct correlation between the consumer GeForce products and the either the Quadro or Tesla lines of professional cards.  That is clearly the case where the Quadro line seemed to take forever to integrate the GT200 products to any successful degree.  The Tesla SKUs are much more complicated and NVIDIA points out that all they have done is disclose one or two upcoming SKUs, not the whole lineup.  I was “guaranteed” that there would be more SKUs “with 512 shaders.”

NVIDIA also specifically stated that the GF100 products that are due out next year are not going to be the same as the Tesla products discussed at the supercomputing conference and that “there will be 512 (shader) parts on both sides.”  What would be different between the two products would be WHEN the 512 options were introduced.  It sure seemed like NVIDIA was trying to say that they would have a consumer-based 512 shader GF100 part when the GeForce lineup is revealed without just telling us

UPDATE: NVIDIA Fermi continues to have problems according to multiple sources - Graphics Cards  3
NVIDIA CEO holds up the first Fermi reference card – though not really…

The NVIDIA rep also took issue with us calling Fermi products late; company CEO Huang stated on a November 7th conference call that Fermi would be available in Q1 2010.  Obviously March 2010, as the rumors suggest, are in Q1 2010, if just barely.  So sure, we have to give NVIDIA credit that PUBLICLY they have been saying all the right things since November about the product release.  Of course, the original goal for Fermi’s launch was in late September at the NVIDIA-hosted GPU Technology Conference.  Instead, we were treated to an architectural overview of the technology that left us drooling quite a bit.  Unofficially, at the show, I heard from several people that the goal was to have Fermi available for CES. 

Well, with that show just 7 days away, and the March 2010 rumors circulating, that obviously isn’t going to happen.  So while publicly, and to investors (which is what matters most to NVIDIA), Fermi has only been delayed once.  But for me, it has been delayed at least once more. 

So while NVIDIA wants to refute the three main claims of our original story (Oak Ridge fail, shader count fail and delay fail), I think there are lot of questions that need to be addressed before we’ll get the truth.  Unfortunately we’ll never know for sure what happened internally with Oak Ridge and NVIDIA (could be nothing, could be major) or how much internal struggle and delay there was at NVIDIA with the Fermi release dates.  We will know for sure (hopefully by March 2010) whether or not NVIDIA can deliver the 512 shader processors they promised us in September.