You Can Buy a Modded RTX 2080 Ti With 22GB of VRAM For $500

A range of RTX 2080 Ti graphics cards have appeared on online listings with double their original VRAM quantity, as VideoCardz reports. With 22GB of GDDR6 and a $500 price tag, AI developers have a new option for their inferencing rigs. In a world of $5,000 RTX flagship graphics cards and shortages of just about everything, modders are getting creative with their creations and side-project income streams.

Getting a new graphics card with enough VRAM to run some of the latest AI open source models is difficult and expensive. Top Nvidia GPUs like the RTX 5080 and 5090 are between 50% and 100% higher than their original launch prices, and though AMD cards are cheaper, they lack CUDA support for Nvidia tools.

While the RTX 2080 Ti is a nearly eight-year-old card, it launched with 11GB of VRAM, and modders have found a way to double that capacity while keeping the card working well. To do it, they replaced the original 1GB GDDR6 chips with 2GB chips from more modern cards.

Credit: Zhou’s store on Ebay

“Nvidia GeForce RTX 2080 Ti Turbo with 22GB GDDR6 VRAM modified, optimized for AI mapping applications,” one listing reads. They do suggest they can’t guarantee the brand of card you receive, but that it will have 22GB of VRAM and is great for accelerating AI computing tasks.

It also says it’s good for gaming, which I would love to test. The RTX 2080 Ti in its original form is roughly as powerful as an RTX 5060 or RX 9060 XT, with its mere 4,352 CUDA cores, but it does have over 600GB/s of memory bandwidth in its standard configuration.

With double the VRAM and upscaling, it would probably be a decent 1440p or even light 4K gaming card in 2026. It does have Tensor and RT cores for DLSS and ray tracing, but lacks the frame generation support of newer cards. Fortunately, that’s where lossless scaling can step in.

Still, I wouldn’t buy one of these cards for gaming. They’re very old and well out of warranty. You aren’t likely to get much of a guarantee with these home-modded cards, either.

For gaming, a 9060 XT with 16GB of RAM is a much better choice. But for AI workloads? That’s a lot of VRAM for not a lot of money. It might be worth considering.