The stories surrounding NVIDIA’s coming graphics architecture Fermi have been many and the rumors have taken another spin as SemiAccurate and Charlie Demerjian has published an article that paints a gloomy pictures of NVIDIA’s coming circuit. Charlie that is know for his battles with NVIDIA may not be the most unbiased, but the well written articles brings up several interesting points that all is meant to point to that NVIDIA’s Fermi architecture is in such big trouble that the first generation will be nothing but a big black hole in NVIDIA’s wallet.
Charlie points out that NVIDIA’s Fermi architecture is now six months late and that in spite of the long development time there are problems that haven’t been solved. The enormous graphics circuit will very expensive to make and when the quality of the dies is extremely low it simply adds up to an impossible equation. Once again, bring out the salt shaker, even though Charlie says his sources are engineers at NVIDIA.
Beside the die size, 550mm2, that is said to be nearly impossible to cooperate with TSMC’s 40nm technology there is also problems with the power consumption, which not only makes it expensive to own but also very hot.
If they were to go in and solve the problems, that are on the transistor level, they would have to rebuild the architecture, but that would have brought new problems due to the even larger circuit and higher power consumption if they don’t tone down the performance.
GeForce Fermi 100
The article is very much a doomsday prediction where it says NVIDIA won’t be launching anything but PR products in 2010. Fermi II at 28nm us said to be the solution, but there is a long way to go before we’re there.
How much, if any, substance there is in the article is hard to say, but we do know that NVIDIA have been working uphill and against the wind for a long time and there is probably a reason why we haven’t heard anything more about test samples or deadlines for the new architecture.
NVIDIA’s partners have told us that they know very little about the final cards and when they will actually launch. At CeBIT next month NVIDIA is expected to reveal a bit more, but perhaps not as much as we would like.
