Whoop of Ass Tegra?
Windows on Arm Tegra would be my guess.
iMacmatician
Regular
From SemiAccurate: Exclusive: Nvidia has two Keplers in house, but not the big one
A non-GPU Fermi/Kepler hybrid GK117, a low-end GK107, and no GK100 in sight (Charlie says mid-2012 at the earliest for that one).
I wonder what NVIDIA will have as a high-end until then (assuming this report is true). Continue with GTX 580? A GTX "585" that's a higher clocked version of the 580? A lower-end-than-GTX 590 dual-GPU card (< 300 W too) made from low-power 560 Ti's or presumably its 28 nm successor (assuming the latter comes out early enough).
A non-GPU Fermi/Kepler hybrid GK117, a low-end GK107, and no GK100 in sight (Charlie says mid-2012 at the earliest for that one).
I wonder what NVIDIA will have as a high-end until then (assuming this report is true). Continue with GTX 580? A GTX "585" that's a higher clocked version of the 580? A lower-end-than-GTX 590 dual-GPU card (< 300 W too) made from low-power 560 Ti's or presumably its 28 nm successor (assuming the latter comes out early enough).
It's amazing that he doesn't even consider the possibility that chips can be taped out without his knowledge - the guy has reached a whole new level. If there will indeed be 28nm shrinks of high end Fermi then I fully expect them to serve as midrange 6xx parts on the desktop. Not sure what Kepler brings to the table but GF204 on 28nm can't be much worse than a potential GK104.
DegustatoR
Legend
There should not be any Fermis on 28nm. Unless plans have changed again.
That seems quite likely. I think NVIDIA figured that the usual top-to-bottom strategy would make them miss the Ivy Bridge generation of laptops, so they went bottom-to-top instead.
Of course, under such circumstances, one can't help but wonder how long it will take them to launch GK100.
caveman-jim
Regular
In time for Windows 8?
What is your expectation of the Win8 launch?
party @ Arm hq
Yeah, Office365 seems to be rooted for that.
I got bad feelings afterall....You know about the Power consumption of FermiHave a look in your notebook
The X2090 looks like it sits on a glorified MXM module.
And why shouldn't it work for the GPU, when CPUs are also cooled this way?
Well, but Power 7 whose TDP rase to 900W+ Also use this way = =, as well as NVIDIA has a unique program setting for restricting power.
Glad to see they know releasing a smaller chip at first.
But, "Unfortunately, there is no news about the more exciting Kepler chips, and there are no signs of them appearing in GeForce GPUs any time soon"
So funding is still a bit up in the air. Otherwise nvidia wouldn't talk about "up to" 18,000 GPUs. And they obviously won't replace the 960 Tesla M2090 cards (the total amount of GPU slots in Titan is 19,200).
That they don't talk about that earlier 30 PFlop figure but only the "potential to deliver over 20 petaflops peak" tells me, the peak DP Flops of a Kepler GPU is <=1.4 TFlop/s.
Anyway, that upgrade with Kepler GPU is planned to start H2/12 either way. Still plenty of time.
From those numbers alone, you could arrive at a mere 0,9 TFLOPS DP for each Tesla, if I did my math right (18000, 85% of peak including some M2090s alright).
But: You say they only mention potentially over 20 PFLOPS peak as if it where something new. It isn't at least since last years SC.
But: You say they only mention potentially over 20 PFLOPS peak as if it where something new. It isn't at least since last years SC.
