To me, the frequency has become a "blob". It means nothing. If anything, I see more frequency and I think more power draw and less stability.
Real performance is a function of IPC and frequency and that is the only thing that really matters.
That seems reasonable across microarch versions, but presumably is totally reasonable within a product-release-cycle? i.e. to compare the FX-9590 to an FX-8350 based off clock speed. If it's the same core count and microarch, presumably clock is about the only difference, aside from other features like PCIE lanes, cache, etc.
Sure, but the performance doesn't scale linearly with the frequency. So how can we account for that? Also, the i9 10900F and i9-10850K have the same turbo. How long can they sustain in for? On how many cores? What frequency is left afterward? This gets even weirder with more recent technologies that push the frequency higher depending on power delivery stability and thermal constraints.
I agree; the frequency is indicative of performance within a product lifecycle. But I find it too meaningless to give an approximate delta on the performance. And this assumes the process you are running is not too much memory bottlenecked or I/O bound and this delta in performance can be measured. (And the setup can sustain higher thermals).
While not perfect, I prefer to look up benchmarks that resemble my workload and call it a day. Processors are too complicated for me to try to compare with simple numbers.
Because I have seen quite a few consumers get guided too much by the "more frequency, more better" motto.
Moreover, it doesn't stop there. I have seen papers where the only thing cited is the frequency of the processors, and no other reference to the manufacturer/architecture/version of the processors used.
> IPC itself is somewhat arbitrary. in the real world, IPC constantly varies with the mix of instructions, cache-friendliness of the program/data, memory performance, etc.
Absolutely, I would also be skeptical of an "IPC" metric. Performance is the only thing that counts.