Chinese Factories Push GPU Boundaries: From RTX 4090 Mods to Rumored 128GB RTX 5090

$142 upgrade kit and spare modules turn Nvidia RTX 4090 24GB to 48GB AI card -- technician explains how Chinese factories turn gaming flagships into highly desirable AI GPUs

Remember those Frakenstein GeForce RTX 4090 48GB graphics cards emerging from China? Russian PC technician and builder VIK-on has provided detailed insights into how Chinese factories are transforming the GeForce RTX 4090, once regarded as one of the best graphics cards, to effectively double its memory capacity specifically for AI workloads. As a mainstream product, the GeForce RTX 4090 does not support memory chips in a clamshell configuration, unlike Nvidia's professional and data center products. Essentially, this means that the Ada Lovelace flagship only houses memory chips on one side of the PCB. In clamshell mode, graphics cards typically feature memory chips on both sides of the PCB. This limitation is addressed by the GeForce RTX 4090 4GB "upgrade kit," which sells for around $142 in China. The upgrade kit comprises a custom PCB designed with a clamshell configuration, facilitating the installation of twice the number of memory chips. Most components are pre-installed at the manufacturing facility, requiring the user to solder the GPU and memory chips onto the PCB. Additionally, the upgrade kit includes a blower-style cooling solution, designed for integration with workstation and server configurations that utilize multi-GPU architectures. VIK-on demonstrated the process of extracting the AD102 silicon and twelve 2GB GDDR6X memory chips from the MSI GeForce RTX 4090 Suprim and installing them onto the barebone PCB. The technician utilized spare GDDR6X memory chips from defective graphics cards, thereby obtaining additional GDDR6X memory at no cost. Clearly, this operation requires specialized soldering skills and access to appropriate high-end tools. The technician also uploaded a leaked, modified firmware onto the GeForce RTX 4090 48GB. It is important to note that each graphics card possesses a unique GPU device ID, which contains all pertinent information. During the system initialization process, the firmware verifies whether the GPU device ID corresponds with the one embedded within the chip. Hacked firmware has been present for some time. Indeed, it was during the era of the GeForce RTX 20-series (Turing) that enthusiasts uncovered the capability to deactivate memory channels. This feature was not advantageous for the general public, as it was illogical to impair a fully functional graphics card by reducing its memory capacity. However, for repair professionals, this discovery proved invaluable, enabling them to salvage graphics cards with defective memory channels. Consequently, this led to the emergence of unorthodox models in the market, such as the GeForce RTX 3090 with 20GB of memory instead of the standard 24GB, or the GeForce RTX 3070 Ti with 6GB of memory instead of the expected 8GB. The firmware modders identified the possibility of expanding memory capacity through the modification. Consequently, the GeForce RTX 4090 48GB and the GeForce RTX 4080 Super 32GB came into existence. Upgrading the GeForce RTX 4090 to 48GB is an expensive endeavor. First, it is necessary to possess the graphics card in order to extract the Ada Lovelace silicon and GDDR6X memory chips. If you do not have any GDDR6X modules readily available, you'll need to purchase each module, currently priced at $24 on Chinese e-commerce platforms. Consequently, the total cost for the upgrade is $430, excluding shipping costs. Assuming you were fortunate enough to purchase a GeForce RTX 4090 at its original MSRP of $1,599, the total amounts to $2,029. These GeForce RTX 4090 48GB graphics cards typically sell for around $3,320 in China, so you're saving close to 39% - again, assuming you have the soldering skills and access to all the equipment necessary for the upgrade. Alternatively, you can pay someone more qualified to perform the upgrade for you. RTX 4090 supply has already started to dwindle, meaning that sooner or later, Chinese factories will likely begin experimenting with the GeForce RTX 5090, if they haven't already. A rumor is already circulating about the GeForce RTX 5090 128GB. While it may seem like a scam now, it could become a reality further down the road. Hunger is a bitch, and with the U.S. limiting China's access to high-end AI GPUs, it's just a matter of time before the next gaming flagship is t

Upgraded Nvidia RTX 5090 gets 128GB VRAM and $13,000 price tag -- 'super limited' GPU is described as a 'prototype'

We've seen our fair share of modified GPUs from China -- a region restricted from receiving most high-end cards. These export controls imposed by the The White House have created an underground market where gaming-grade graphics cards are stripped of their internals and put onto custom PCBs designed for AI workloads. AI requires a lot of memory, and despite former flagship GPUs like the RTX 4090 already coming with large memory buffers, it's still not enough for tasks such as running LLMs locally. The latest to join the line of modded GPUs is the RTX 5090 with a whopping 128 GB of VRAM. Yes, this graphics card most likely has more memory than your entire PC. It's described as a prototype because the modder is using GDDR7X memory that is not yet available on a larger scale, but there's still a "super limited" quantity associated with this, which suggests that more units could be made. By default, the RTX 5090 comes with 32GB of GDDR7 memory, which might seem plenty, but clearly there's room for improvement. What's interesting is this mod is a tier above anything we've seen before, or thought was possible. Previously, an RTX 4090 was upgraded with 96 GB of VRAM by using the best GDDR7 modules available in a dual-sided config. That was an anomaly, too, as the usual modded 4090s top out at about 48GB VRAM. The reason behind which is the complexity of the custom PCBs that are needed to support such tampering. Currently, the highest-density GDDR7 chips are 24Gb (or 3 GB) per module, which can allow for up to 48 GB of VRAM. To reach higher capacities like 96 GB, dual-sided PCBs are used with memory modules on top and bottom, as seen on Nvidia's RTX Pro 6000. Therefore, for an RTX 5090 to feature 128 GB of VRAM, the factory would either need access to unreleased 32 Gbit GDDR7 chips -- which no vendor has announced yet -- or design a Frankenstein custom PCB with far more memory chips than usual; 64 GB on either side, totaling 128 GB in a dual-sided config. Not only that, but custom firmware and BIOS are also required for said VRAM to be properly recognized and utilized by your system. The screenshot in the source tweet is from Nvidia-SMI which clearly shows the full 128GB memory being detected, meaning the mod works and is not just proof-of-concept. Moreover, the fact that there's pricing for this card is even more evidence that it's a potentially commercially-viable product ready to be sold. According to I_Leak_VN, the 128 GB 5090 is priced at roughly $13,200 which is more than 4x the street pricing of a regular 5090. Shockingly, it almost makes sense considering the 96 GB RTX 6000 Pro sells for $10,000. If you do the math, it's 33% more memory for 33% more price with the souped-up 5090. Also, none of that accounts for the actual "craftsmanship" of performing the mod by hand. Regardless of the illicit nature of these GPUs, someone has to work diligently to make them possible, and it's not easy to casually swap GPU cores and VRAM chips. Unfortunately, after the core and memory is extracted from a normal 5090, they often make their way onto secondhand marketplaces where they're sold as real 5090s to scam buyers. It's important to stay vigilant, especially now that the AI boom has created this new precedent where various hollow 4090s and 5090s have flooded the market for cheap, so if anything looks to good to be true, it likely is.

NVIDIA GeForce RTX 5090 128 GB GPU Spotted: Custom Memory, Designed For AI Workloads & Priced At $13,200 Per Piece

NVIDIA's GeForce RTX 5090 GPU has started to receive the custom memory implementation with up to 128 GB capacity for a massive price. Chinese AI Factories Now Creating Custom NVIDIA GeForce RTX 5090 GPUs With Up To 128 GB Memory Capacities Factories in China have been dedicated to producing custom AI GPUs based on the NVIDIA GeForce RTX lineup. These factories spawned as soon as the AI boom started a few years ago. We have seen cards such as the RTX 4090 with 48 GB and even up to 96 GB VRAM (though that turned out to be a one-off thing), up from its standard 24 GB configuration. Now, it looks like the RTX 5090 is going to receive the custom memory treatment. How these factories achieve such mods is by salvaging the GPU and memory from gaming graphics cards such as the GeForce RTX 5090, and equipping them to custom PCBs, which often allow for dual-side memory configurations. This allows them to create custom graphics cards with much larger capacities. It isn't just a simple component swap, as the proper firmware and BIOS are also used to ensure compatibility. Now, memory modding isn't the only thing that happens in these factories. Gaming GPUs also get converted into two-slot blower-style designs for ease of use in server and workstation setups. According to I_Leak_VN, it looks like the latest model to get this treatment is the NVIDIA GeForce RTX 5090. In stock, the card features 32 GB of GDDR7 memory, but the custom variant features a massive 128 GB frame buffer, quadrupling its memory capacity. While we don't have more specifics or pictures of the PCB itself (yet), it looks like the card is made through a combination of higher-capacity GDDR7 memory dies, which would lead to 64 GB on a single-sided PCB. The dual-sided PCB layout doubles the memory to 128 GB. Now there's a question to raise here, which memory modules is this card actually using? We know GDDR7 is available in 16 Gb and 24 Gb densities, which enable 2 or 3 GB memory per module, so 32 or 48 GB. Even a dual-sided model would end up with 96 GB VRAM, which is the same as the RTX PRO 6000. So, unless this factory was able to source 32 Gb density modules out of nowhere (currently, no memory vendors offer these), it does look like this could either be a hoax or some weird Frankenstein of a mod is going on with the PCB. In the screenshot provided by the source, the card appears with its full 128 GB memory, and is running the 550.144.03 driver. The card is also given a price. The standard NVIDIA GeForce RTX 5090 retails for $1999 US (MSRP), but is more realistically found around $2500-$3000. The custom 128 GB variant is reportedly priced at around $13,200 US. That's a 6.6x higher price for 4x the memory capacity. While the price is quite hefty, it makes sense that the card is priced so high. The RTX PRO 6000 Blackwell with a 96 GB VRAM capacity costs around $10,000 US. In terms of price, the RTX 5090 128 GB variant is 33% more expensive and offers 33% more memory than the PRO Blackwell. Once again, the RTX PRO 6000 is an official variant offered by NVIDIA, while the 128 GB GeForce RTX 5090 is a custom mod that doesn't have any approval from NVIDIA. It's sold via the factories themselves to AI customers since that much VRAM has no practical or sensible use in gaming at the moment. The popularity of these custom VRAM designs is surprisingly big in Asian markets, which are deprived of proper AI GPUs due to their high prices (including tariffs) and trade restrictions over certain regions. We will dig for more information on this model.