P-Core and E-Core always 100% usage

[Else00]

M3 Max 40 Core GPU

[SUPERCELLEX]

Same system, same problem
It's reporting 100% CPU... at 0.5 watts

[tydia]

same here

[aliasaria]

confirmed that parse_cpu_metrics() doesn't recognize the cores properly on an M3 Max.

For example on an M3 Max with 14 cores (4 Efficiency, 10 Performance) the E-Cluster_active is set to 100 but the the actual cores are named E-Cluster0_active, E-Cluster1_active to P-Cluster13_active

Some sample code that is able to read all cores is:

        for cluster in e_core:
            cpu_metric_dict["E-Cluster_active"] += cpu_metric_dict["E-Cluster" +
                                                                   str(cluster) + "_active"]
            cpu_metric_dict["E-Cluster_freq_Mhz"] = max(
                cpu_metric_dict["E-Cluster_freq_Mhz"], cpu_metric_dict["E-Cluster" + str(cluster) + "_freq_Mhz"])

        cpu_metric_dict["E-Cluster_active"] = int(
            cpu_metric_dict["E-Cluster_active"]/len(e_core))

        for cluster in p_core:
            cluster_name = "P-Cluster" + str(cluster) + "_active"
            cluster_active = cpu_metric_dict[cluster_name]
            cluster_freq = cpu_metric_dict["P-Cluster" +
                                           str(cluster) + "_freq_Mhz"]

            if cluster_freq != 0:
                cpu_metric_dict["P-Cluster_active"] += cluster_active

            cpu_metric_dict["P-Cluster_freq_Mhz"] = max(
                cpu_metric_dict["P-Cluster_freq_Mhz"], cluster_freq)

        cpu_metric_dict["P-Cluster_active"] = int(
            cpu_metric_dict["P-Cluster_active"]/len(p_core))

But even the code above doesn't look the same as the OSX Native activity monitor. The last P-Cores show high usage in asitop and powermetrics on the second performance cluster (P-1 Cluster, CPU 9-13) because these cores can be online or offline and when they are offline they may report high usage which skews the results. So we have to filter out cores that are not online but I am not sure how.

[aliasaria]

Looks like 5314e67 may be the necessary fix

[tydia]

Looks like 5314e67 may be the necessary fix

Yes, or simply use gavi's fork fixed the issue

pip install git+https://github.com/gavi/asitop.git

[ufo2243]

same here

[ufo2243]

same here

pip install git+https://github.com/gavi/asitop.git works for me