The NVIDIA GeForce RTX 5090 currently ships with 2 GB GDDR7 modules, but even 3 GB modules have seen limited adoption due to their high cost. However, that could change soon. According to Weibo leaker Golden Pig Upgrade, memory manufacturers are reportedly planning 4 GB and 6 GB GDDR7 chips for a 2027 to 2028 window, a jump that could reshape how much VRAM budget and mid-range GPUs can carry without needing a wider memory bus. A 128-bit card using four memory packages currently tops out at 8 GB with today's 2 GB chips. Move to 4 GB chips and that same card could hit 16 GB, or 24 GB with 6 GB modules. Scale that up to a 256-bit interface with eight packages, and you're looking at 32 GB or 48 GB respectively, without touching the PCB layout or bus width. Of course, going the other way is possible too, a narrower memory bus combined with higher density, faster memory chips, can lead to similar bandwidth, with lower relative cost.

This would matter most at the low end, where narrow 128-bit buses have made generous VRAM capacities either awkward or expensive to pull off. Videocardz, citing supply chain sources, notes that today's 3 GB GDDR7 chips are already unusually expensive due to extreme demand, and that cost has reportedly contributed to delays around the rumored RTX 5050 9 GB and the RTX 50 Super refresh. If 4 GB chips end up cheaper to produce at scale, they could become the more practical choice for mainstream cards, while 6 GB modules might make more sense for workstation and other high-memory GPUs first. It's worth noting that the leaker's timeline only points to sometime in 2027 or 2028, so there's no confirmation this lines up with NVIDIA's RTX 60 series or AMD's next Radeon generation specifically.