NVIDIA’s China-focused RTX 4090D graphics card based on the AD102 processor, designed to circumvent US export restrictions, is now available with double the memory capacity of 48 GB.
The new graphics cards were reported by insider @bdsqlsz, who gained access to a cloud computing platform where these devices were already installed. The main use of these graphics cards is to train large language and other generative artificial intelligence models that require not only significant computing power but also memory.
While the original tweet doesn’t include a photo or physical evidence, it does confirm the RTX 4090D’s memory bandwidth at 937 GB/s, which is the same as GDDR6X memory speed. The 48GB capacity is available with the AD102 GPU, which is commonly found in RTX ADA workstations but uses the GDDR6 standard. This indicates that the 48GB RTX 4090D has faster memory. In addition, the software lists a GeForce RTX 4080 SUPER 32 GB, which usually has 16 GB of GDDR6X memory.
Memory modifications like this are not uncommon, especially in the modding community and GPU service centers where memory replacement is a daily task. However, the fact that a cloud computing platform offers such modified instances of graphics cards indicates a more systematic use of them.
The RTX 4090D normally supports only 12 memory modules, but it can be used with the RTX 3090 Ti, which supports 24 modules, along with the RTX 4090D’s AD102 GPU. This solution allows you to double the amount of memory. It is likely that Chinese servers use such modifications.
Sources: Tom`s Hardware, VideoCardz
Spelling error report
The following text will be sent to our editors: