If you’ve ever been left frustrated by the mere 24GB of VRAM in an RTX 4090D or found the pathetic 16GB attached to an RTX 4080 Super to be utterly below you, there could be a solution available, if you’re resourceful. That’s because one Chinese Cloud AI company has been shown using these GPUs but with double their normal VRAM allocation. So, if you ask really, really nicely, they might sort you out with a new double VRAM Nvidia graphics card too.
Nvidia isn’t set to release any new best graphics card contenders until the much anticipated RTX 5000 series launch at the tail end of this year or start of next year. However, it seems that hasn’t stopped this unnamed company from finding its own way to eke out even more performance from its Nvidia cards.
The leaked information comes via X user @bdsqlsz who posted on the service several screenshots of systems running what claim to be RTX 4090D and RTX 4080 Super GPUs with double their normal amount of VRAM. That means the RTX 4090D is sporting 48GB while the RTX 4080 packs in 32GB.
The cards are being used by a Chinese Cloud computing service where the extra memory is useful for AI computing applications, rather than it being used to store gaming data.
Notably, the PCB that houses the RTX 4090D GPU isn’t normally one that’s able to accommodate enough memory chips to be able to offer 48GB of VRAM, as it only houses the chips on one side of the board. Instead, the company is likely to have used old RTX 3090 Ti PCBs that have memory locations on both sides of the board, with these boards also known to be compatible with RTX 4090 GPUs. However, with no pictures of the cards or other information to go on, we can’t know for sure.
You may be wondering why a company like this would go to such efforts to upgrade an RTX 4090D (or indeed the RTX 4080 Super) when the RTX 4090 is more powerful and there are available even more powerful dedicated server GPUs that can inherently be equipped with more VRAM.
Answering those in reverse order, using consumer-grade chips like the RTX 4090D and RTX 4080 Super is just much cheaper than buying the server-grade options. Meanwhile, the RTX 4090 isn’t an option for use in China at the moment due to US export restrictions. Indeed, that’s why the RTX 4090D exists – it’s a cut-down variant of the 4090 that’s approved by the US government for use in China. This has led to many Chinese firms having to become extra resourceful to find the powerful cards they need.
Realistically, your chances of getting hold of a card like this are very slim, as it’s fairly unlikely such a company would be willing to sell you one, and getting the product shipped from China might be tricky too. You could endeavor to do the mod yourself or employ some other modder to take on the task, but if your aim is to improve gaming performance, it’s an expensive, risky option we wouldn’t recommend. You’ll get basically no performance improvement in games, as almost no game max out the 24GB of memory on an RTX 4090D as it is.
Instead, you’re much better off just splashing the cash on a full-fat RTX 4090, for a start. Or, if you’re actually looking for a card that offers decent value while also providing blazing performance, the RTX 4080 Super is the card to go for.