drunkenmaster, i think nvidia could still surive if they played the game right, they really need to court Apple, real hard.
Apple won't settle for just onboard okish FPU performance on a crappy intel gma plugged next to their cpu die.
They are gonna want to have bells and whistles attached, nVidia could provide that, Apple users are already willing to pay more for the name. Apple's marketshare is increasing a lot lately.
Well Apple got boned by Nvidia on the bumpgate thing with a huge failure rate and Nvidia wouldn't take the blame, as per usual. Not long ago Apple bought the entire remaining supply of 4850/70's AMD had in stock at that time, which was something like 5-10k cards because they wanted a reliable, cheap and powerful card for their high end computers.
Secondly, on cpu die is, as I tried to explain going to massively increase, even in Intel's utterly crap current intergrated gpu, you're talking about a HUGE massive increase in FPU power, compared to every cpu of the past 5 years it will absolutely blow them away.
But keep in mind that the intergrated is only set to vastly improve. AMD's first intergrated is set to be a 480shader lower end varient of a 5870(so a 5450 or something along those lines). You're talking about, I dunno, a 9600gt/9800gt's worth of gpu power on die. But thats current cpu's, when you get up to 16/32 cores in a couple years, if you need FPU over Int performance, you'll just buy a varient with a higher ratio of gpu/fpu units over int units. Intel within a few years will have very very decent gpgpu types cores to intergrate.
Even if Nvidia got a contract to supply all of Apple's computers, its not a huge number, Apple aren't a massive supplier compared to the rest of the OEM's put together, infact they are still rather tiny and utterly dwarfed by normal PC sales. Once you cut of 80% of your sales you can't afford to invest the same amount in developing your gpu, if they cut spending theres every chance their gpu's and gpgpu's in the future simply won't be competitive.
THe main problem being Nvidia will have no way to compete with Intel and AMD on manufacturing price, they can't afford to open their own fab, they will never match the savings from building them inhouse like the other two can.
The fact a company is so desparately courting such a ridiculously small amount of their own market(gpgpu) tells you they know theres trouble ahead.
I think its bad news for the short term, Larabee isn't ready, Nvidia have point blank refused to deal with the manufacturing issues that have plagued them for 2 years and keep running headlong into design problems due to large core design. If they don't change that in the next chip well, I can't see it being anything but just as late as Fermi. You can't build a chip that big on such a small process anymore, its just not doable.
The latest news is that Nvidia's biggest design win for a gpgpu supercomputer they announced on that launch day with the fake Fermi a couple months ago has been canceled, because that partner says Fermi is too hot, to late and not fast enough to warrant it.
