I use a 4K external display which must be driven by the NVIDIA GPU, and I'm seeing the performance degradation all the time on High Sierra. The whole OS is very jittery and laggy: resizing windows, switching tabs in web browser, etc... Having looked through a few forums, it seems like quite a number of people are also experiencing performance issues using the same or similar MacBook Pro (mid 2014) with the NVIDIA GeForce GT 750M GPU Card.
I realise that these benchmarks may be something of a gimmick but "my UI is jittery, youtube videos aren't playing as smoothly as before" is quite vague and there are quite number of explanations for those symptoms. I'm trying to find some repeatable way to demonstrate that the problem exists, and also hopefully some way of quantifying the effect on performance. That way it should be easy to get everyone with the same problem to file bug report/support requests that are actionable.
So... I've just taken the plunge and switched back to Sierra to run some tests. Graphics performance is back to normal, and everything is running smoothly - so it doesn't look like a hardware issue (at least not for me).
In the spirit of attempting to compare "like for like", I ran the same GPU benchmark on the same laptop on Sierra 10.12.6: https://browser.geekbench.com/v4/compute/1359623. From a quick scan of the results:
* High Sierra gets better numbers for the tests that it can run (Metal2 is supposed to bring improvements over Metal).
* High Sierra can't run one of the tests (though Sierra can run it). Given that these tests are normally designed to cover a wide range of functionality between them, this could indicate that there is some feature implementation missing for this GPU on High Sierra, or the implementation isn't sufficient to run the test.
So not absolutely conclusive, but it does show the difference between the GPU performance on the two systems.
I'm interested if anyone has any other ideas of tests/benchmarks that might help narrow this down. Now that I can switch between 10.12.6 and 10.13 it's relatively easy for me to try things on both and compare. If we can find a definite way to show that it's an issue, it should be much easier for Apple to fix. My support requests and bug reports have not gotten me anywhere so far but I'm going to add this latest benchmark and hope for the best!