Show "instructions per clock" for E- and P-cores
samuelfredrickson opened this issue · 5 comments
powermetrics reports this metric for each CPU core cluster, as well as an average for the entire SoC. I've found it interesting to see how this value fluctuates under various workloads, and think it would be neat to see it incorporated into asitop.
What does this value show exactly? Is it better higher or lower? I find it very difficult to interpret.
@NightMachinery It is a metric of how many instructions we executed by the cpu cluster, per clock cycle, during the metric sample.
Higher is better, assuming you are looking for higher cpu performance (from faster execution of instructions).
@BitesPotatoBacks Some instructions do more work than others though. Perhaps not as much of an issue in ARM CPUs.
@NightMachinery Of course, certain instructions may take multiple cycles to complete (but these timings are often tied to the efficiency of the microarch). Though, I do not know the exact timings of the instructions supported by current Apple Silicon.
IPC is an average number for the cluster that gives the metric. Longer execution times for instructions would decrease the value.
This is a very interesting metric, Im pretty sure you can also get this metric on ryzen cpu's so it would be very interesting for comparisons between different CPU's