Nvidia's entry-level graphics card lineup is about to get strange. According to hardware leaker MEGAsizeGPU, the company is preparing a refreshed RTX 5050 that swaps 8GB of GDDR6 memory for 9GB of GDDR7, but at a cost: the memory bus shrinks from 128-bit to just 96-bit.
On paper, this looks like the kind of trade-off nobody asked for. A narrower bus usually spells trouble for bandwidth. But here's where it gets interesting. The new GDDR7 chips run at 28Gbps instead of 20Gbps on the existing GDDR6, meaning the RTX 5050 would actually gain about 5% in memory bandwidth, from 320GB/s to 336GB/s. You get an extra gigabyte and faster data pipes, just with an odd configuration that would have seemed unthinkable a few years ago.
The remaining specs stay put. The card will use the same GB207 die with 2,560 CUDA cores, built on TSMC's 5nm process, rated at 130W.
Why this matters for budget cards
The move reveals something about today's memory market. The main reason for this change is reportedly GDDR6 supply and demand conditions, rather than a simple upgrade in capacity. In other words, GDDR6 is harder to get hold of. Rather than offer a straightforward 12GB RTX 5050 on a 128-bit bus, Nvidia apparently decided to use three 3GB GDDR7 chips instead of four 2GB GDDR6 ones. The math works, but it's the kind of engineering constraint that usually stays hidden in a data sheet.
Budget GPUs don't typically command much attention, but this one highlights real supply pressures rippling through the industry. The RTX 5050 is one of the few GPUs that basically saw no price hikes in the past few months, only going up about $10. The only other card with the same stability was the RTX 5060. These cards have become a relative bargain, which makes Nvidia's need to tinker with their specs all the more telling.
The RTX 5060 gets a bigger surgery
Things get messier with the RTX 5060. Nvidia is working on a new RTX 5060 with a cut-down version of the GB205 GPU, the silicon that powers the RTX 5070. Apparently, Nvidia has told partners to focus on the 8GB RTX 5060 Ti, which has led to a shortage of GB206 dies for RTX 5060 SKUs.
In practical terms, this means Nvidia would be repurposing GPUs meant for higher-end cards just to keep the RTX 5060 in the pipeline. The RTX 5070 has a 12V-2x6 connector and the GB205 die inside features 6,144 CUDA cores, those would be reduced to 3,840 CUDA cores for an RTX 5060. Moreover, the bus width on the GB205 is 192-bit and that would also go down to 128-bit, in order to match an RTX 5060.
Board partners will need to design new printed circuit boards to accommodate the GB205 die while keeping the 8-pin power connector RTX 5060 buyers expect. These non-standard GPU variants are often limited to certain markets, so do not expect a global or major launch. It would more likely be something you stumble upon when reading VideoCardz or checking reviews from Chinese outlets.
The pragmatic read
None of this is glamorous. Nvidia isn't unveiling innovative architectures or breakthrough features. Instead, we're seeing a company working through a memory crisis by shuffling components around the bottom of its product stack. The RTX 5050's 96-bit bus would be the first time Nvidia has used such a narrow interface since the RTX 3050 in 2024. The RTX 5060 variant is basically Nvidia making do with what it has.
For consumers, the practical impact is minimal. Budget GPU buyers care about price and performance, not whether their card uses a 96-bit or 128-bit bus. Both new configurations should deliver familiar performance levels for the entry-level market. What these moves really signal is that even Nvidia, despite its market dominance, isn't immune to supply chain headaches. When component constraints get tight enough, even the world's largest GPU maker has to get creative.