Chinese developers create modified 48GB Nvidia RTX 4090D and 32GB RTX 4080 SUPER GPUs for the AI cloud

Chinese companies develop custom modded RTX 4090D and RTX 4080 SUPER GPUs for the AI Cloud server market

Companies in China have been caught using custom/modded Nvidia GeForce RTX graphics cards for their AI cloud servers. This included memory-doubled RTX 4090D and RTX 4080 SUPER graphics cards with 48GB and 32GB of GDDR6X memory respectively.

Why do these GPUs exist? The simple answer is AI. Nvidia’s GeForce RTX gaming GPUs are popular for enabling cheap access to large amounts of AI compute. That said, running large AI models require a large amount of computation power and memory capacity. This is why companies are modding Nvidia’s GeForce RTX GPUs to give them larger memory pools to work with.

Memory modifications for graphics cards are not uncommon. However, they are usually done by GPU modding enthusiasts and as part of GPU repairs. Now, large cloud computing platforms are using memory-modded RTX 40 series graphics cards in high volumes. These GPUs may use custom GPU PCBs to enable support for larger memory capacities. This suggests that there are large GPU modding operations in China. However, it’s unknown what methods were used to create these custom GPUs. After all, there are not pictures of these custom RTX GPUs.

(48GB RTX 4090D and 32GB RTX 4080 SUPER graphics cards – via @bdsqlsz)

The supply of custom/modded RTX GPUs in China is making high levels of AI compute more readily available within the country. This makes Nvidia’s sanction-busting RTX 4090D graphics card even more appealing in China. After all, 48GB of memory is typically only available in ultra-expensive workstation grade graphics cards.

You can join the discussion on Nvidia’s memory-modded RTX GPUs within Chinese AI servers on the OC3D Forums.