It has been reported that some Chinese companies running servers dedicated to intensive computing and AI are modifying RTX 4090D and 4080 Super graphics cards to boost their performance. These “tinkerings” to RTX 4090D and RTX 4080 SUPER graphics cards enable memory to be doubled, increasing these versions to 48 GB and 32 GB GDDR6X memory respectively. The high-end RTXs can run elaborate AI models, but memory quickly becomes their bottleneck. These “boosted” cards are intended for training large linguistic models and other generative AI models requiring considerable computing power and memory capacity.
It is with these modifications that China can bypass barriers and expand its AI computing capabilities in the country. Under these conditions, Nvidia’s sanctions-bypassing RTX 4090D graphics card becomes even more attractive in China. Indeed, in order to have access to 48 GB of memory, you logically have to equip yourself with expensive professional cards, which are above all subject to the embargo. Modifications therefore become relevant on condition that you have access to PCBs and technical capabilities… and just as well, China is home to the majority of graphics card manufacturers’ factories.
