The explosion of AI workloads has created an odd side effect: graphics cards that seemed outdated a couple of years ago are suddenly back in demand. The reason is simple. Running large language models, image generators, and 3D rendering software often depends less on raw GPU horsepower than on how much video memory is available. That’s the gap a small Japanese repair shop called AiGPU is stepping in to fill.
AiGPU specializes in hardware repair and modification, and it has started offering a service that physically reworks NVIDIA’s GeForce RTX 2080 Ti, pushing its VRAM from the original 11GB up to 22GB. The price for the upgrade sits at roughly $282, or about 45,000 yen, and it’s aimed at owners who already have the card and want more headroom for memory-hungry software.
The process isn’t a simple firmware tweak. Technicians actually desolder the card’s standard 1GB GDDR6 memory chips and swap them out for 2GB modules soldered directly onto the circuit board. That alone wouldn’t work, though, because the GPU needs to be told how to recognize and address the new memory. So the shop also modifies the memory configuration resistors and rewrites the card’s BIOS, allowing the graphics processor to properly manage the expanded memory pool without crashing or misreading capacity.
For customers who’d rather skip the DIY hassle of sending in their own card, AiGPU also sells pre-modified RTX 2080 Ti units that already come with the full 22GB of memory installed.
The 2080 Ti isn’t the only card getting this treatment. AiGPU has extended similar memory-boosting mods to several other popular NVIDIA models, giving budget-minded buyers and small studios more options for stretching aging hardware further than NVIDIA ever intended.
There’s an important caveat here, though. Doubling the memory doesn’t touch the GPU’s actual processing power, and it doesn’t widen the memory bus either. So gamers hoping a 22GB 2080 Ti might crank out better frame rates than the stock 11GB version are likely to be disappointed — the modification isn’t a performance upgrade in that sense at all.
Where it actually pays off is in workloads that choke on limited VRAM rather than limited compute. Workstations running large language models, AI image and video generation tools, or complex 3D rendering pipelines frequently hit a wall not because the GPU is too slow, but because there simply isn’t enough memory to hold the data being processed. A modified 2080 Ti gives those systems breathing room, turning what used to be an outdated gaming card into a surprisingly cheap alternative to buying new professional-grade hardware.
For studios and hobbyists working on projects that used to stall out or crash entirely due to memory limits, that extra 11GB could mean the difference between a job that finishes overnight and one that never finishes at all. Whether enough people are willing to send their cards to Japan — or buy an already-modified unit — to make this a mainstream fix remains to be seen, but it’s a clever reminder that old silicon still has some life left in it when the right tinkerer gets involved.
