Two 16Gb (2GB) part numbers in Micron's GDDR7 catalog are now listed as "End of Life" (EOL). VideoCardz and others reported the change on September 22, and when we checked the public catalog on September 23, both the 28GT/s and 32GT/s versions showed EOL. The 16Gb chips have been found in some GeForce RTX 5060 cards, so component sourcing options could narrow. But a part number going EOL is not the same as a card being discontinued. Looking at the status of Micron's remaining 24Gb (3GB) parts alongside the design of existing 8GB cards also shows which figures are still missing for judging the impact.
Where the four part numbers stand in the catalog
Micron's GDDR7 part catalog lists four part numbers in total: 16Gb and 24Gb, each at 28GT/s and 32GT/s. In the current catalog, both 16Gb parts are EOL, and only the 24Gb 28GT/s part is in Production. The 24Gb 32GT/s part is listed as "Contact Sales," which does not mean it is in production. Here is the status as listed on September 23, by density and speed.
| Capacity per chip | 28GT/s | 32GT/s |
|---|---|---|
| 16Gb (2GB) | MT68A512M32DF-28:A: End of Life | MT68A512M32DF-32:A: End of Life |
| 24Gb (3GB) | MT68A768M32DF-28:A: Production | MT68A768M32DF-32:A: Contact Sales |
16Gb and 24Gb refer to how much data a single chip can hold, while 28GT/s and 32GT/s refer to the per-pin transfer rate. These are separate axes. Even with the 16Gb 32GT/s part going EOL, the 24Gb 32GT/s part is not automatically a mass-production replacement, because its catalog status is Contact Sales rather than Production.
The table shows that this is not Micron withdrawing from GDDR7 as a whole. The scope is two 16Gb parts, at two speeds, going EOL. The move to 24Gb also differs by speed. Summing it up as "all 3GB parts are in production" would misread the status of the 32GT/s part.
Micron's product brief describes 16Gb, x32-configuration parts with 28Gb/s and 32Gb/s part numbers. The current catalog, by contrast, shows product status. The fact that the brief still describes the 16Gb parts cannot be taken as proof that they remain in regular production. Conversely, the EOL label alone does not reveal the last order date, the last shipment date, or the amount of inventory in distribution. We also could not find a published EOL notice or a designated successor part.
What happens to 8GB cards when 2GB chips go away
The confirmed example of adoption is the GALAX GeForce RTX 5060. VideoCardz reported that this card, spotted in February this year, carried four 16Gb chips from Micron, identified as D8FHL. The current EOL reports likewise trace back to a discovery by UNIKO's Hardware and the change in Micron's catalog. What can be confirmed is the configuration of certain cards, not that Micron parts were used across all RTX 50 models or all production lots.
According to NVIDIA's official specifications, the RTX 5060 carries 8GB of GDDR7 on a 128-bit memory bus. The part numbers in Micron's catalog are x32 per chip, meaning a 32-bit interface. Four 32-bit chips make 128 bits, and four 2GB chips make 8GB. The chip count VideoCardz reported is consistent with this specification.
So can the 3GB parts that Micron lists as in production simply be swapped in? Using four of the same 32-bit-wide parts would, by calculation, give 12GB. That capacity differs from the current RTX 5060's 8GB specification. Both densities have 266 balls in the catalog and the same 12.00 × 14.00 × 1.10 mm package, but a matching footprint does not mean they can be dropped unconditionally onto an existing card. Compatible memory control, board design, operating conditions, and a change to the product specification may need to be validated. At the very least, Micron listing the 24Gb part as Production does not mean NVIDIA has announced a 12GB RTX 5060.
EOL does not point to a supply crisis
What the EOL label directly shows is the product lifecycle status of two Micron part numbers. The catalog does not tell us when existing inventory or already-contracted shipments will run out. The share of Micron's supply that went to RTX 50 has also not been disclosed. So from this listing alone, we cannot calculate any change in GeForce RTX 50 production volume, retail stock, or prices.
The effect on sourcing will depend on which cards and which lots had qualified Micron parts. The VideoCardz example of the GALAX RTX 5060 shows that Micron was not merely a candidate on paper. At the same time, that single case cannot establish that Micron was a major supplier, nor that its volume was small enough to have no effect. How much Samsung and SK hynix, the companies named in the reports, can substitute also cannot be judged without knowing their production capacity and the board makers' qualification status.
Switching to 3GB parts would also mean a change in capacity and product specification for 8GB cards. That has to be considered separately from buying additional chips of the same capacity from another manufacturer.
What to watch next
First, we need the last-order and last-shipment terms Micron sets for the two 16Gb part numbers. If NVIDIA's and the board makers' qualified part numbers, adoption ratios, and inventory periods are added to that, the timing and scope of how the component EOL reaches card production can be narrowed down. For Micron's 24Gb parts, the turning point will be how the 32GT/s part moves from "Contact Sales" and whether products that actually use it appear.
For the 8GB RTX 5060, the EOL of the 2GB part does not immediately promise a move to 12GB. Still, if the capacity per chip goes up while the same 128-bit configuration is kept, a card's capacity design could change. Only when Micron's product status, NVIDIA's official capacity specification, and the part numbers actually mounted on cards all move together can we say this component change has altered the options for consumer cards.