Memory manufacturers are reportedly planning 4GB and 6GB GDDR7 memory chips for a 2027 to 2028 window, a jump from the 2GB modules on today’s NVIDIA GeForce RTX 5090 and the 3GB modules that have seen only limited adoption because of their cost.
The claim traces back to Weibo leaker Golden Pig Upgrade, reported by TechPowerUp. Nothing here is confirmed by NVIDIA, AMD or any of the three memory makers, and a two-year runway leaves plenty of room for these plans to change before a single chip ships.
How denser GDDR7 memory chips change VRAM math
GDDR7 capacity is set by two things: how many memory packages sit on the board, and how much each package holds. A 128-bit graphics card carrying four packages tops out at 8GB with today’s 2GB chips. Swap in 4GB chips and the same board reaches 16GB; swap in 6GB chips and it reaches 24GB, with no change to the PCB layout or the width of the memory bus. Scale that up to a 256-bit card with eight packages and the ceiling moves to 32GB or 48GB respectively.
The trade also runs the other way: a narrower bus paired with denser, faster chips can match the bandwidth of a wider one at lower relative cost. That is where this matters most, at the budget end, where a 128-bit bus has made generous VRAM capacities either awkward or expensive to pull off without a costlier board.
Micron already off the 2GB tier
Micron has discontinued its 2GB GDDR7 chips, the same modules used across several GeForce RTX 50 series cards. Both the 28Gbps D8FHL part (MT68A512M32DF-28:A) and its 32Gbps sibling are now listed as end of life on Micron’s own site, leaving 3GB modules as the only GDDR7 capacity Micron still lists.
Micron only started supplying GDDR7 to NVIDIA in meaningful volume earlier this year and was never the primary source for RTX 50 cards; Samsung and SK Hynix still cover the bulk of that supply, so the change should not cause an immediate shortage on its own. It does say something about where Micron would rather put its manufacturing capacity right now. The 3GB chips that remain are already unusually expensive, driven by demand that has reportedly contributed to delays around the rumoured RTX 5050 9GB and the RTX 50 Super refresh.
What the RTX 60-series timing might mean
The 2027-to-2028 window for denser GDDR7 lines up awkwardly with NVIDIA’s own roadmap slipping rather than speeding up: teqpost reported that the GeForce RTX 60-series reportedly slips to 2028. If that holds, the first cards with a real shot at launching with 4GB or 6GB modules from day one would be the 60-series, not a refresh of the current RTX 50 line, which will already be several years into its 2GB and 3GB chips by the time denser packages exist in volume.
What to watch
Whether Samsung and SK Hynix follow Micron in retiring their own 2GB lines, and whether the 3GB squeeze eases enough for the rumoured RTX 5050 9GB and RTX 50 Super refresh to actually ship, are the nearer-term signals worth tracking. The 4GB and 6GB chips themselves remain two to three years out and unconfirmed by any of the three memory makers.
