Collect synchronized power and thermal telemetry

Create the telemetry logger

The logger finds hwmon devices and channels by name and label. The use of both name and label keeps the script working if Linux assigns different hwmon directory numbers after a reboot.

Create collect-telemetry.sh in ~/perf-per-watt, replacing ampere_hwmon and thelio_hwmon with the devices for your system:

    

        
        
cat > collect-telemetry.sh <<'EOF'
#!/usr/bin/env bash
set -eu

output_file=${1:-telemetry.csv}
interval=${2:-1}

find_hwmon() {
    wanted_name=$1

    for device in /sys/class/hwmon/hwmon*; do
        if [ "$(cat "$device/name" 2>/dev/null)" = "$wanted_name" ]; then
            echo "$device"
            return 0
        fi
    done

    return 1
}

find_labeled_input() {
    directory=$1
    sensor_type=$2
    wanted_label=$3

    for label_file in "$directory"/"$sensor_type"*_label; do
        [ -e "$label_file" ] || continue

        if [ "$(cat "$label_file")" = "$wanted_label" ]; then
            echo "${label_file%_label}_input"
            return 0
        fi
    done

    return 1
}

to_watts() {
    awk -v value="$1" 'BEGIN { printf "%.6f", value / 1000000 }'
}

to_celsius() {
    awk -v value="$1" 'BEGIN { printf "%.3f", value / 1000 }'
}

ampere_hwmon=$(find_hwmon apm_xgene) || {
    echo "The apm_xgene hwmon device was not found" >&2
    exit 1
}

thelio_hwmon=$(find_hwmon system76_thelio_io) || {
    echo "The system76_thelio_io hwmon device was not found" >&2
    exit 1
}

cpu_power_file=$(find_labeled_input "$ampere_hwmon" power "CPU power")
io_power_file=$(find_labeled_input "$ampere_hwmon" power "IO power")
soc_temp_file=$(find_labeled_input "$ampere_hwmon" temp "SoC Temperature")
cpu_fan_file=$(find_labeled_input "$thelio_hwmon" fan "CPU Fan")
intake_fan_file=$(find_labeled_input "$thelio_hwmon" fan "Intake Fan")

printf '%s\n' \
    'timestamp,epoch_seconds,avg_freq_khz,cpu_power_w,io_power_w,soc_power_w,soc_temp_c,cpu_fan_rpm,intake_fan_rpm' \
    > "$output_file"

while true; do
    timestamp=$(date --iso-8601=seconds)
    epoch_seconds=$(date +%s.%N)

    avg_freq_khz=$(awk '
        { total += $1; count++ }
        END {
            if (count > 0) {
                printf "%.0f", total / count
            } else {
                print "0"
            }
        }
    ' /sys/devices/system/cpu/cpufreq/policy*/cpuinfo_cur_freq)

    cpu_power_w=$(to_watts "$(cat "$cpu_power_file")")
    io_power_w=$(to_watts "$(cat "$io_power_file")")
    soc_power_w=$(awk -v cpu="$cpu_power_w" -v io="$io_power_w" \
        'BEGIN { printf "%.6f", cpu + io }')
    soc_temp_c=$(to_celsius "$(cat "$soc_temp_file")")
    cpu_fan_rpm=$(cat "$cpu_fan_file")
    intake_fan_rpm=$(cat "$intake_fan_file")

    printf '%s,%s,%s,%s,%s,%s,%s,%s,%s\n' \
        "$timestamp" \
        "$epoch_seconds" \
        "$avg_freq_khz" \
        "$cpu_power_w" \
        "$io_power_w" \
        "$soc_power_w" \
        "$soc_temp_c" \
        "$cpu_fan_rpm" \
        "$intake_fan_rpm" \
        >> "$output_file"

    sleep "$interval"
done
EOF

chmod +x collect-telemetry.sh

    

The script samples every second by default. One-second sampling captures sustained changes without spending excessive time reading sysfs files.

Test the logger

Start the logger in the background, collect five samples, and stop it:

    

        
        
sudo ./collect-telemetry.sh telemetry-test.csv &
logger_pid=$!
sleep 5
sudo kill "$logger_pid"
wait "$logger_pid" 2>/dev/null || true

    

Inspect the first records:

    

        
        
head telemetry-test.csv

    

The output is similar to:

    

        
        timestamp,epoch_seconds,avg_freq_khz,cpu_power_w,io_power_w,soc_power_w,soc_temp_c,cpu_fan_rpm,intake_fan_rpm
2026-08-06T09:47:56-05:00,1786027676.633482916,1109099,12.240000,8.025000,20.265000,36.000,1065,720
2026-08-06T09:47:57-05:00,1786027677.708797530,1031262,12.280000,8.024000,20.304000,36.000,1065,735
2026-08-06T09:47:58-05:00,1786027678.783871620,1018438,12.360000,8.024000,20.384000,36.000,1065,735
2026-08-06T09:47:59-05:00,1786027679.863988207,1025325,12.360000,8.025000,20.385000,36.000,1065,720
2026-08-06T09:48:00-05:00,1786027680.940551795,1006262,12.360000,8.025000,20.385000,36.000,1065,735

        
    

Your values will differ. Confirm that each row has nine fields and that power, temperature, and fan readings are nonzero.

Count the number of fields in each row:

    

        
        
awk -F, 'NF != 9 { print "Unexpected field count on line", NR, ":", NF }' telemetry-test.csv

    

No output means every row has the expected number of fields.

Interpret the recorded values

The logger records the following values:

The average frequency is a system-wide summary. It doesn’t show whether individual CPUs run at different frequencies, but it’s sufficient for comparing OpenSSL runs using all cores.

What you’ve accomplished and what’s next

You’ve now created and tested a CSV telemetry logger.

Next, run OpenSSL while the logger records frequency, power, temperature, and fan speed.