https://www.eteknix.com/teclab-makes-its-own-nvidia-2080-ti-super-gpu-video/
 
Following the release of a video, however, the team at TecLab has decided to take matters into their own hands. In essentially butchering 3 graphics cards, they have been able to essentially create their very own Nvidia 2080 Ti ‘Super’ GPU.  As you may be aware, the essential (and most basic) difference between the standard and ‘Super’ varients of the Nvidia 20XX range is the memory. As such, the team at TecLab decided that if they could put some of the new ‘Super’ memory into an existing 2080 Ti, then it would effectively make the GPU a ‘Super’ model. A nice and solid idea… in theory. It is, however, certainly more than a little terrifying in practice!
 
So, how did they achieve this? Well, by stealing the 16 Gbps modules from two GALAX RTX 2080 SUPER graphics cards they inserted these into a 2080 Ti replacing the existing 14 Gbps memory. Yes, this is about as terrifying as it sounds! Amazingly, however, through a lot of highly intricate soldering, they were not only able to do this but also to get it working and as you might expect, with the utilization of an OC Lab water block, they were able to get some pretty amazing benchmark results!
 
Although they were able to achieve higher speeds, the team eventually stuck at 2250 MHz for optimum stability. With this providing 17200 MHz effective speed, however, this is 3200 more than the stock 14 GHz provided by the base 2080 Ti and, as such, represents a huge ‘leg-up’ in comparison to the base model performance.
 
The video in the article needs to be translated so turn the subtitles on should you wish to watch it. 

 
But super cards also come with more cuda cores. =)

Sajin

But super cards also come with more cuda cores. =)

We can simulate performance of a lower model with a higher clocks, it's just math: multiply the core with base clock and raise memory bandwidth to scale.


Edit a day later, for example:

An RTX2080Ti Super will most likely have the extra CUDA cores of an RTX Titan along with faster Vram. So raise the clocks on RTX2080Ti to base of 1547MHz, boost of 1874MHz, and Vram to 15.27Gbps and at least the rasterization performance ought to match-- up until the details and resolution is raised to the point where the 24GB is more useful.

Not counting Tensor cores in this simplified calculation.