Modified GeForce RTX 4080 with 32 GB flood the Chinese graphics card market - prices around $1480

The modified GeForce RTX 40 market in China looks like it's heating up again. A growing number of GeForce RTX 4080 cards—originally 16 GB GDDR6X—are showing up on second‑hand sites with 32 GB advertised. Sellers are asking north of 10,000 yuan (≈ $1,480), which is almost three times the price of recently reviewed RTX 2080 Ti units that were expanded to 22 GB.

These are not new SKU releases but standard RTX 4080 PCBs altered to present extra memory to the GPU. Stock RTX 4080 ships with 16 GB of GDDR6X on a 256‑bit bus and 9,728 CUDA cores. Simply doubling the DRAM capacity does not widen that 256‑bit bus or boost raw memory bandwidth. The tweak reportedly involves swapping or adding memory chips and reconfiguring the board so the GPU enumerates 32 GB. Use case? Mostly AI/ML inference and other tasks where fitting a larger model in local VRAM matters more than peak bandwidth.

Recent listings suggest the practice has moved beyond lone modders and repair shops. Standard RTX 4080s are being converted in larger quantities, enough that a niche in the used‑GPU market has emerged—locally referred to as Goofish.

There are obvious trade‑offs. These mods aren’t supported by NVIDIA or the AIBs, manufacturer warranties are void, and users may need patched drivers. Driver issues can be the trickiest part: some buyers report reapplying patches whenever they switch to gaming. By contrast, the 22 GB RTX 2080 Ti conversions didn’t seem to suffer the same driver headaches. Long‑term reliability will hinge on the PCB, how the memory was soldered or installed, and who did the work.

Bottom line: if raw VRAM size is the only metric that matters, a 32 GB‑flashed RTX 4080 gives more local memory than a stock RTX 4090’s 24 GB—and costs a lot less than workstation GPUs with 32 or 48 GB. But it’s a compromise: extra GBs for less bandwidth, no official support, and some operational risk.