Question Area vs Frequency for "5nm" cores

[Geddagod]

 

[Geddagod]

Disclaimer: Haven't done a sanity check yet, so if there are any obvious mistakes, feel free to point them out. Plan on doing that, as well as adding more cores, later.
Chart:
Red: TSMC (N4P, N4, N5 etc etc)
Green: Samsung (5LPE, 4LPE, 4LPP, etc etc)
Blue: Intel (Intel 4)
Data labels : core + product, for example, A510 E23 = A510 in the Exynos 2300, A715 G1 = A715 in the Tensor G1, X2 M90 = X2 in the Mediatek 9200
Core Area in mm2, Fmax in GHz
Core Area:
Core + L1 for cores with shared L2 cache, for cores with private L2 cache, we subtract L2 SRAM + tag arrays. Cores with shared L2s have a large advantage even with trying to compensate by removing the L2 arrays since the control logic still takes up a large amount of space.
For example, a Zen 4 core using the method we describe above is 2.81mm2, however if we remove the entire L2 "unit" from the cores we end up with 2.52mm2. I'll add a disclaimer for which cores have a shared vs private L2 cache later on the graph.
Fmax:
The peak frequency that each core hits according to spec from the companies themselves, whatever I can find on notebookcheck, and other sources. Sometimes Kurnal labels the peak frequency on the die shots themselves.
Cores in desktop or laptop form factors gain an obvious advantage over their mobile competition.
Especially for the little cores, the frequency listed might not be the Fmax they can hit when pushed to the extreme.
As for the decision to use Fmax rather than perf in general to include IPC in the comparison, finding perf measurements for the little cores seems to be hard, don't see many places that test them. A chart could prob easily be created for the P-cores though, even of the smart phone chips, by using a common bench like GB5 or GB6.

Just something I thought was interesting.