NVIDIA’s GeForce RTX 2080 Ti, released in 2018 with 11 GB of GDDR6 memory, has been doubled in VRAM capacity thanks to hardware modifications. These used graphics cards, with their memory upgraded to 22 GB, are being offered for sale at a price of $499, especially for users working with local AI models.

In this process, carried out by Hong Kong-based sellers, the original 11 GB memory chips of the cards are replaced with higher capacity modules. Checks performed using GPU-Z software show that the modified cards recognize the 22,528 MB VRAM capacity without any problems.

Technical Details and Applications of Modified Cards

The graphics cards offered for sale mostly consist of models with blower-type cooling systems suitable for server and workstation use. While buyers cannot necessarily choose a brand, depending on stock availability, RTX 2080 Ti cards from different manufacturers such as ASUS, MSI, GIGABYTE, or Leadtek are sent.

Listed on eBay for $499, these products offer an alternative, especially for those who want to run diffusion models used in large license plate models and image creation processes locally. 22 GB VRAM provides a valuable advantage in high memory-required processes where the model or processed information needs to fit into the graphics card memory.

The RTX 2080 Ti, thanks to its Tensor Core units within the Turing architecture, can handle these types of AI workloads. However, since the card’s architecture is older than current NVIDIA GPUs, it does not offer the performance advantages of current cards in newer generation low-precision data formats or heavy diffusion workloads.

Price Advantage and Market Position

From a pricing perspective, these modified 22 GB RTX 2080 Ti models stand out as a more affordable option compared to other NVIDIA solutions with similar memory capacity. Used Titan RTX models with 24 GB VRAM sell for around $800, while Quadro RTX 6000 models sell for around $900.

A more current option, the RTX 3090 with 24 GB of VRAM, costs approximately $1,200, making it considerably more expensive than these modified cards. Following export restrictions imposed on high-performance NVIDIA GPUs, there has been a noticeable increase in the number of these VRAM upgrade processes carried out in workshops in China.

According to user feedback, these cards can be actively used to run models such as the Qwen 32B. What do you think about these modified graphics cards?