Metal

Another whole 2 stream processors/cores/whatever they're deciding to call them these days, which results in like a 25% increase in the # of vector execution lanes (CUDA cores in NVIDIA marketing speak).

It's also got 3GB more memory, but that likely doesn't really matter to you. The faster memory on the 2080 super would be better for gaming workloads than the extra capacity on the 2080 Ti. That said, there are other workloads that definitely could utilize it, but if you're doing that you probably just want an HBM2 card like the titan or radeon vii anyways (but radeon vii is EoL... so unless you need it for passthrough...).

Basically, it can do more math at once.

There also may be some hardware difference in the "tensor core" stuff, but since I've only ever worked on the code for the open-source AMD drivers, I really don't know how that hardware is laid out, or how the machine code that runs on it even works.

The 2080 ti use’s gddr6 that is slightly faster then the super’s gddr6

@Revise is actually right from a second look. So, there's a memory bandwidth bump as well.