RTX 4090 with 48 GB VRAM and water cooling – Chinese manufacturer undercuts RTX 5090 on price

While NVIDIA does not officially have a GeForce RTX 4090 with 48 GB of memory in its portfolio, recent discoveries from China show that this very card has already reached the market – including a factory-fitted all-in-one (AIO) water cooling system. And at a price that is lower than what you currently have to pay for the RTX 5090 in the wild. A curious example of market-specific developments that NVIDIA prefers to keep officially under wraps.

48 GB GDDR6X – Twice the VRAM at twice the price? Not quite.

According to Task Manager and GPU-Z, the GPU is based on the familiar AD102-300 silicon, which is also used in the regular RTX 4090 with 24 GB GDDR6X. With 16,384 CUDA cores, an identical 384-bit interface and a memory bandwidth that does not increase due to the same clock rates as the standard model, the card remains technically largely the same. The only difference is the doubling of the video memory, which here is a whopping 48 GB. In this form, the memory is likely to be designed for AI workloads and data-intensive applications in particular – gaming advantages are hardly to be expected given the limited gaming engines. The decisive indication of the intended use therefore lies in the memory itself, not in the computing power.

BYKSI AIO cooler – triple fan plus radiator

The card shown was apparently equipped with a customized water cooling system from BYKSI, a China-based provider of custom water cooling solutions. In addition to the radiator, three axial fans are mounted on a massive cooling block that covers the GPU including voltage converter and memory chips. The lettering “GeForce RTX 4090 48 GB” is emblazoned on the packaging as well as on the radiator itself, which at least visually gives the impression of an officially approved product – even though NVIDIA has no official authorization here. The design is more reminiscent of professional solutions in the workstation sector than a typical consumer product, which once again underlines the area of application: data centers, AI training or industrial rendering processes.

Pricing policy: 48 GB cheaper than 5090?

With a rumored price of around 3400 US dollars, the card is above the MSRP of an RTX 4090 (MSRP: 1599 USD), but below what is currently being asked for an RTX 5090 – if you can get one at all. The RTX 5090, also based on the AD102, is still hard to find, and AIO versions in particular sometimes cost significantly more than the USD 3400 of the 48 GB version. However, this price advantage is deceptive: if you are purely interested in gaming, you will still be better off with the 24 GB version or the 5090. The additional storage capacity of the 48 GB version only reveals its potential in professional applications where data volumes beyond the 24 GB limit need to be processed.

96 GB version in the works – The next step for the AI market

Alongside this 48 GB version, information is already circulating about an upcoming 96 GB version of the RTX 4090, also with a focus on the AI market. Initial images and benchmark data are still missing, but given the rapid demand from the AI sector, such a card would no longer be a surprise. Whether this is a converted consumer card or a dedicated board design is currently unclear.

Conclusion: NVIDIA remains silent – the Chinese market delivers

All in all, this discovery shows once again how strongly the Asian market is decoupling itself from Western standards. What NVIDIA does not include in the official lineup is quickly implemented by local OEMs themselves. The fact that these variants exist at all and are sold in limited quantities is a sign of a changing GPU market: less consolidation, more niche solutions – especially for memory-intensive special applications. Whether and when NVIDIA itself will follow suit with an official RTX 4090 with expanded memory remains to be seen. Until then, Chinese manufacturers will remain the driving force for unconventional solutions.

Source: CeFurkan via Reddit