The “China Edition” NVIDIA GeForce RTX 5090D is a standard RTX 5090 SKU with firmware modifications.

It is anticipated that NVIDIA’s next GeForce RTX 50 series will launch at CES 2025 in January. Nonetheless, NVIDIA is present in several regions, including the Chinese market, where it has a substantial foothold. In order to comply with export laws that prohibit Chinese companies from purchasing GPUs capable of training AI models, the company has produced its “Dragon” version of the top-tier GPU versions for Chinese players. Only the GeForce RTX 5090D has been confirmed thus far, thanks to MEGAsizeGPU, although NVIDIA seems to be working on “D” versions for both the RTX 5090 and RTX 5080 cards. The RTX5090D appears to have the same hardware specifications as the standard RTX 5090, according to a recent leak from the Chinese Chiphell forum.

The RTX 4090 and RTX 4090D were distinct in the previous generation, with the RTX 4090D having a lower TGP and core count. But this time, NVIDIA will simply use firmware to lock specific functionalities, physically leaving the same anticipated GB202 “Blackwell” die with CUDA core count and memory. It is anticipated that the RTX 5090D would have a lower Tensor core capabilities and maybe a lower frequency of the entire chip because the main objective of the export restrictions the US is enforcing is to limit Chinese access to GPUs capable of training and inferring AI models. In addition to potentially slowing down NVIDIA’s AI technologies like DLSS, this might prevent some apps from leveraging these Tensor cores for inference.

The RTX 4090 and RTX 4090D were distinct in the previous generation, with the RTX 4090D having a lower TGP and core count. But this time, NVIDIA will simply use firmware to lock specific functionalities, physically leaving the same anticipated GB202 “Blackwell” die with CUDA core count and memory. It is anticipated that the RTX 5090D would have a lower Tensor core capabilities and maybe a lower frequency of the entire chip because the main objective of the export restrictions the US is enforcing is to limit Chinese access to GPUs capable of training and inferring AI models. In addition to potentially slowing down NVIDIA’s AI technologies like DLSS, this might prevent some apps from leveraging these Tensor cores for inference. While other general-purpose AI operations are restricted, NVIDIA might manage to let DLSS and other gaming-related technologies function regularly. During the crypto mining boom, readers might remember Low Has Rate (LHR) cards, which featured a software lock that was comparable to cap mining hash rate.