Brazilian overclocking team TecLab has created one powerful Frankenstein of a GPU, dubbed “RTX 4090 ‘SUPER’“. Of course, NVIDIA never made a GPU with this name (though there were attempts at making an even more powerful model), but the reason behind this naming likely has to do with a common tactic NVIDIA applies to SUPER cards: by adding faster VRAM.
RTX 4090 “SUPER” Uses Different PCB, Overclocked GDDR6X
However, this isn’t just a simple VRAM mod, as TecLab have to borrow two GPUs to build this overclocking monster. The first donor is the GALAX GeForce RTX 3090 Ti HOF OC LAB Edition, which is designed to break overclocking records – here, the original silicon is removed and the PCB is taken off to pair it with the AD102 silicon that powers the RTX 4090. Next, the RTX 4080 SUPER’s 23Gbps GDDR6X modules are extracted to give the flagship silicon more memory bandwidth to play with.
Once the parts are swapped, paired, and configured properly, the team starts out by measuring the performance on Unigine Superposition’s 8K benchmark, and begins overclocking from there. A simple VRAM overclock that brings GDDR6X modules’ effective bandwidth to 26Gbps has net a 13% increase in scores, while a 3GHz GPU overclock on top of it has increased the gap to 16%. Finally, cranking everything to the max with tweaked LOD settings has raised the scores up to 39% over stock figures.
While this likely wouldn’t count towards any world records, it does prove that RTX 4090 certainly can do with faster memory, especially in high-resolution scenarios. With next-gen RTX 5090 expected to use GDDR7 modules (with 28Gbps modules), this should feed the flagship Blackwell GPU with huge amounts of bandwidth for a huge increase in performance over today’s Team Green flagship.
Source: Wccftech
Pokdepinion: If RTX 5090 indeed is using 28Gbps, what’s stopping modders from acquiring 32Gbps modules and go from there? Would be interesting to see how fast that’ll go.
