According to several test sites (one at the end as an example), while one can't really o/c the core speed, if you o/c the Infinity Fabric speed as well as memory (memory, only to a point, then surprisingly it slows down), you can get a good boost -- see https://www.youtube.com/watch?v=Ssuqhyqah2k 

Also, new, expensive MSI X570 Motherboard with all the bells and whistles -- https://www.guru3d.com/articles-pages/msi-meg-x570-godlike-review,22.html.

 

Yes, I mentioned earlier that these high frequencies don't necessarily scale too well. When the system shows a drop off or a flat spot that could mean some part of the system has reached a limit. Scaling beyond can also mean decreased performance overall.

Let's simplify it really badly, even so I think can still be a good representation of what's going to happen:

Say we have one architecture with 10 instructions per cycle and another with 8.

To complete the instructions requires filling the data values or parameters. With the latest CPUs they have a number of sub processors that are constantly analysing the code for branch prediction. This leads to predicting the memory pages that will be required to obtain the values in time. That's complicated by which cores are being utilised, and by how many can concurrently be supported.

You might find your 8 core running cores at 35% produces a smoother simulation using six cores running 60%.

When we get an out come we might find that the 10IPC chip completes on average 5 instructions, and the 8IPC chip averages 6 instructions.