Hardware Slop?: While repairing a nonworking piece of gaming hardware is a common occurrence, adding extra RAM during the process is still unusual. However, a Chinese modder recently said that this kind of modification has now become widespread among enthusiasts.

In China, specialized modders can add significantly more AI acceleration capabilities while repairing broken Blackwell GPUs. A technician claimed that turning standard GeForce RTX 5080 cards into 32GB monsters is already "a thing" in the country. If confirmed, the news could have some interesting consequences for the high-end GPU market.

The GeForce RTX 5080 was originally introduced in January 2025, sporting a standard VRAM size of 16GB. So far, Nvidia is only providing 32GB with the biggest and baddest Blackwell model (RTX 5090), though the company is rumored to introduce a GeForce RTX 5080 Super model with 24GB of VRAM soon.

According to more recent rumors, the alleged new GeForce model could arrive later than expected. The 24GB card may not arrive until the second quarter of the next year, following a soft launch that's internally scheduled for CES 2026. Even if these claims prove accurate, the new GPU will still have much less VRAM than the high-end GeForce RTX 5090 model.

Chinese modders are allegedly working ahead of Nvidia's schedule, though. They are adding 16GB of VRAM to blower-type GPUs, which are designed to eject hot air from the rear. These types of cards are easier to turn into accelerators for AI inferencing and training workloads, expanding the overall computing capacity with multiple cards installed in the same custom server setup.

The Chinese technician who disclosed the new modding/repairing trend said that the GeForce RTX 5080 32GB model is the most popular one. Other blower-style cards from the Blackwell generation (from RTX 5060Ti 16GB to RTX 5070Ti 16GB) are also popular models. The modders are likely exploiting information that emerged from Nvidia's 2022 hacking incident, using custom tools to make repair and upgrade work far easier.

Judging from what I've read in my personal social circles, turning powerful gaming GPUs into yet another AI-focused hardware proposal is not exactly an exciting prospect for everyone. On Reddit, overclocking enthusiasts are highlighting the very sad state of the current RAM market, and the fact that turning even more hardware into "AI nonsense" is not a good modding goal to pursue. In any case, adding 16GB of VRAM to a 16GB GPU is still an interesting type of hardware surgery.