A tech brief from memory manufacturer Micron suggests Nvidia will be adopting GDDR6X memory for at least the RTX 3090, if not other high-end Ampere graphics cards. The news comes from a PDF from Micron (uncovered by Videocardz), which clearly states the next-generation graphics card will be equipped with yet-unannounced GDDR6X rated to 19-21Gbps.
That's seriously speedy memory, and according to the same document [PDF warning] hosted on the Micron website the company expects GDDR6X cards to come equipped with an average of around 12GB of the stuff. Both of which are increasingly important metrics for gaming at high resolution and fidelity. If you crave 4K60, it's not simply a question of moar cores.
For comparison, the RTX 2080 Ti Founders Edition came with 11GB of memory rated to 14Gbps.
This is also the first official word we've received of the GDDR6X memory configuration's existence. Nvidia pulled a similar move for its high-end Pascal graphics cards, which came with shiny and new GDDR5X memory. GDDR5X offered a fairly considerably bump in speed over the GDDR5 standard—10Gbps to 7Gbps, respectively, and later increased to 11Gbps—and the jump with GDDR6 to GDDR6X looks to be no different. That means it may also run with slightly lower voltage demands (and also may be more expensive).
With a price premium likely attached to the faster memory, it's unlikely we'll see GDDR6X proliferate its way down the entire GPU stack.
This document also serves as further proof that Nvidia will be opting for the RTX 30-series branding and naming schema—and even a lofty RTX 3090. Yet Micron doesn't put the cards together, only sells the chips—so there may be some guesswork involved on its part. After taking a second look at the included table, it appears as though the exact RTX 3090 graphics card specification cannot be determined from what's listed.
With an expectation that Nvidia will announce the Ampere generation on September 1, 2020, during its 'GeForce Special Event', it's likely we're hearing the name that's stuck for the GPU generation. The proximity to the event also adds weight to the veracity of the recent specs bumble over at Micron—and I can guarantee that Jen-Hsun Huang won't be happy if someone has spoiled his launch party.