We already know that Kepler, Nvidia's next-gen 28-nm GPU, isn't scheduled to enter production until next year. Well, the folks at SemiAccurate have now gotten some dirt on a batch of other 28-nm parts Nvidia has in the pipeline. Those products are said to be shrinks of existing Fermi GPUs, and their production schedules purportedly suggest Kepler won't hit stores until March 2012 at the earliest.
SemiAccurate shows a list of 10 upcoming 28-nm mobile parts with power envelopes as low as 15-20W and as high as 70-75W. Memory interfaces will range from 64 to 256 bits in width, and they'll be connected to as much as 2GB of GDDR5 RAM on higher-end offerings. Other details are slim; the site says some information is "a bit blurred to protect the exact sub-species of mole involved," so I assume that's intentional.
The most interesting nuggets of information are perhaps the sampling and mass production time frames. Seven of these Nvidia GPUs will start sampling in early October, and those same parts are due to be mass produced in late December. SemiAccurate predicts that Nvidia's Kepler offerings "will trail the above by at least three months, likely more." You do the math.
We'd recommend taking that information with a grain of salt, of course. That said, if SemiAccurate is right, then Nvidia won't have any 28-nm GPUs out this year. AMD, meanwhile, has gone on record to say it will "deliver" 28-nm products this year. We'll just have to see how all of these predictions pan out.
|Rumor: Intel Skylake-X and X299 will headline Computex 2017||48|
|Rumor: Nvidia to answer Radeon RX 550 with GeForce GT 1030||17|
|Samsung Galaxy Book tablets blend Windows 10 and Intel CPUs||16|
|Deals of the week: a mighty PSU, mid-range CPUs, and more||27|
|AMD board partners begin tricking out RX 560s and RX 550s||16|
|Dell shows off a pro-grade 4K HDR display and AIO machines||15|
|Rumor: Google to bake ad-blocking into Chrome browser||51|
|EpicGear's Defiant modular gaming keyboard reviewed||12|
|GeForce cards with faster RAM are inbound from multiple locations||19|
|Those power consumption numbers are very fermi-liar||+53|