What even is this thread? I was looking for a test that shows the IPC of Redwood Cove vs. Raptor Cove, and I stumbled upon this thread.
First of all, considering IPC by merely dividing some aggregate points by an average frequency is criminal, to say the least. Modern day processors are complex in how they boost frequency. Also, you have multiple different clock domains on the chip (e.g., LLCs, NoCs, memory controllers ... etc). The resultant performance, which is easily measured by wall-clock time, is simply the product of all of these domain interacting together to achieve the eventual execution. These domains vary their frequencies to hit some power or thermal goals. Comparing microarchitecture A's IPC to another microarchitecture B's IPC is an intericate task, especially so if these microarchitectures are part of different SoCs using different IPs/sub-systems. However, the best you can do is to _fix_ the frequency of all the different parts of the SoC and execute a benchmark that you can time accuratly. To what frequency you set these sub-system is also important; you do not want to set the frequency for the, say, the LLC too low such that you bottleneck the faster microarchitecture, or too high such that the SoC thermal/power throttle the core. Now, to what frequency you set the other SoC depends on equivalence. If both SoCs use the same organization, then you can match. If not, you need to understand how they work and try to match it. Of course, you can do all of this, **IF** the SoC allows you to set these frequencies. You also need to make sure that the OS scheduler is not interfering and missing up the measurements.
Sadly, I do not have a Meteor Lake CPU on hand for testing. I will see if I can borrow one. But, again, I doubt we can do any apples vs. apples comparision for it, since the mobile SoC has changed from the Raptor Lake one.
