Old Nvidia GPUs Transformed into Affordable AI Accelerators: RTX 2080 Ti Gets 22GB VRAM for $282
·
17.08.2026 19:00
·
2 min read
Japanese enthusiasts have begun offering custom modifications for Nvidia GeForce graphics cards, doubling their VRAM. According to Tom’s Hardware, upgrading the previous-generation flagship RTX 2080 Ti from 11GB to 22GB of GDDR6 memory costs 45,000 yen (around $282).
The process involves desoldering the stock 1GB memory chips and replacing them with 2GB Samsung GDDR6 chips, followed by hardware strap adjustments on the PCB. Factoring in the cost of a used card on the secondhand market (around $250), the completed 22GB accelerator costs the buyer about $500, which is significantly cheaper than specialized RTX Pro series cards.
Practical Benefits: AI vs. Gaming
For traditional gaming, this upgrade is practically useless: the TU102 GPU and memory bandwidth (616 GB/s) remain unchanged, meaning there is no frame rate boost in games.
In local AI workloads, VRAM capacity is the primary bottleneck: 22GB allows users to load large language models (LLMs) and work with long contexts without running into Out of Memory errors. The presence of Tensor Cores in the Nvidia Turing architecture ensures the card remains a budget-friendly solution for AI inference.
Supported Graphics Cards
According to the workshop’s profile on the Mercari marketplace, engineers offer similar memory upgrades for other models:
- RTX A2000: upgrade from 6GB to 12GB;
- RTX 3070: upgrade from 8GB to 16GB;
- RTX 3080: upgrade from 10GB to 20GB;
- RTX 4080 and 4080 Super: upgrade from 16GB to 32GB;
- RTX 4090: upgrade from 24GB to 48GB;
This procedure is an unofficial third-party modification: altering the graphics card’s design completely voids the factory warranty and requires expert modification of the power delivery circuits.
