NVIDIA's "modded" GeForce RTX 4090 with 48 GB VRAM onboard has finally been benchmarked, and the results indicate that the SKU is only suitable for AI workloads.
NVIDIA's RTX 4090 With Higher VRAM Is Only Suitable For AI-Intensive Workloads; Gamers Shouldn't Look Towards It
Well, for those unaware, the markets have seemingly found a workaround to the strict supply situation of AI accelerations, and the solution they have come up with isn't entirely appealing to gamers. Modders have somehow managed to rack up the VRAM count on the GeForce RTX 4090 up to 48 GB in an attempt to make the model perform better in AI-intensive workloads. Now, courtesy of the Russian YouTuber MKA, the modded RTX 4090 has finally been opened, revealing the PCB configuration. He also managed to test out the GPU in some applications.
His particular unit with him was a blower-style GPU with weird NVIDIA branding and a modded-out PCB. By the looks of it, the GPU board was likely custom-made since mounting additional GDDR6X VRAM modules isn't possible with the original design, as it requires support for dual-sided memory. Since this design isn't verified, it is certain to be flawed, and that is what the YouTuber's testing concluded, which we'll talk about next.
While testing out the GeForce RTX 4090 48 GB in synthetic workloads such as Uniengine, the GPU was running at extremely high noise levels, reaching up to 65 dB in some scenarios, which is quite high compared to what a normal RTX 4090 would run at. Despite being in a blower configuration, the temperatures were somewhat acceptable, revolving around the 70°C range and peaking out at 86°C. However, this model is in no way suitable for gaming scenarios, although the performance would be slightly better compared to the normal 24 GB variant.
MKA also ran AI models locally with the GeForce RTX 4090 48 GB in the LM studio, in particular Google's Gemma-2 with 27b parameters, and by the looks of it, the GPU operated without any troubles. AI workloads are the only area where using the modded GPU makes sense, given that with the high VRAM capacity, individuals can carry out intensive tasks without any trouble, and the supply is much better compared to NVIDIA's mainstream AI accelerators.
Such modded GPUs are highly likely to remain confined to Chinese markets, given the high demand for such models. The 48 GB variant is reported to be selling for around $3,400, which is more expensive than the GeForce RTX 5090. So, the demand is definitely there for such units.
About the author: Muhammad Zuhair is a hardware and technology reporter for Wccftech, specializing in the semiconductor industry and the complex interplay between technology, manufacturing, and geopolitics. His coverage focuses on the corporate strategies and technological roadmaps of industry giants like TSMC, NVIDIA, Samsung, and Intel. Zuhair's expertise lies in deconstructing complex topics such as fabrication nodes (e.g., 2nm process), the economic impact of policies like the CHIPS Act, and the strategic development of AI infrastructure from NVIDIA, AMD and Intel.
Follow Wccftech on Google to get more of our news coverage in your feeds.
