Skip to content

Raspberry Pi CPU Frequency: Check Clock Speed and Governors

Run vcgencmd measure_clock arm on the Raspberry Pi to check its current CPU clock in hertz (Hz). A low reading at idle is common. Measure again while the Pi is doing work, then check temperature and throttling flags if the clock stays low.

Read the clock · Compare idle and load · Diagnose a low reading

Raspberry Pi CPU frequency commands

Run on the Pi host:

1
2
3
4
# Firmware Arm clock: hertz.
vcgencmd measure_clock arm
# Linux CPUFreq reading: kilohertz, if this policy exists.
cat /sys/devices/system/cpu/cpufreq/policy0/scaling_cur_freq

Read the number after = in the firmware output. For example, frequency(48)=1800404352 is about 1,800 MHz. Divide the firmware number by 1,000,000 or the Linux number by 1,000 to convert to MHz. To print the Linux reading in MHz directly:

awk '{printf "%.0f MHz\n", $1/1000}' \
    /sys/devices/system/cpu/cpufreq/policy0/scaling_cur_freq

Use the policy limits below to find the range on your board. The firmware and Linux readings come from different interfaces and are sampled at different instants.

Compare idle and loaded clock speed

Start the monitor in one terminal on the Pi and let the readings settle:

1
2
3
4
5
watch -n 1 '
vcgencmd measure_clock arm
vcgencmd measure_temp
vcgencmd get_throttled
'

In a second terminal on the Pi, run the workload that felt slow. On a test Pi, you can also use this eight-second check to see whether the clock responds to demand:

timeout 8s yes >/dev/null

yes repeatedly writes to /dev/null and keeps roughly one CPU core busy. timeout stops it automatically (exit status 124 is expected). Compare the clock before, during and after the workload. Press Ctrl+C to close the monitor. A short run shows how the clock responds to load. For sustained cooling problems, observe your actual workload.

These readings came from a Raspberry Pi 4B running Raspberry Pi OS Trixie with the ondemand governor:

When sampled Arm clock SoC temperature
At idle 600 MHz 34.0°C
Four seconds into the load 1,800 MHz 36.0°C
Five seconds after the load ended 1,800 MHz 34.5°C
Fifteen seconds after the load ended 600 MHz 35.0°C

The same board ran at both 600 MHz and 1,800 MHz without changing its settings. get_throttled stayed at 0x50000, which records earlier undervoltage and throttling. The current flags were clear at each sample. Use your own readings to compare clock behaviour.

Inspect CPUFreq policy before changing it

# Read each policy the kernel exposes.
for policy in /sys/devices/system/cpu/cpufreq/policy*; do
    [ -d "$policy" ] || continue
    printf '\nPolicy: %s\n' "$policy"
    for field in affected_cpus scaling_driver scaling_governor \
                 scaling_available_governors scaling_min_freq scaling_max_freq \
                 scaling_cur_freq cpuinfo_cur_freq; do
        [ -r "$policy/$field" ] || continue
        printf '%s: ' "$field"
        cat "$policy/$field"
    done
done

Each policy controls the online cores listed in affected_cpus. scaling_min_freq and scaling_max_freq give its allowed range in kHz.

scaling_cur_freq often reflects the last frequency requested by the driver. cpuinfo_cur_freq uses hardware feedback when the driver provides it. On the Pi 4B used here, cpuinfo_cur_freq was absent. See the Linux CPUFreq attributes and the firmware clock command.

What the governor changes

Governor, if available Policy intent
performance Request the highest frequency within the policy limit
powersave Request the lowest frequency within the policy limit
ondemand or schedutil Adapt requests to activity using the governor's algorithm
userspace Allow an external controller to request frequency

Actual clocks remain subject to hardware, firmware, power, and temperature constraints. performance is not a promise of an always-maximal measured clock; powersave is not proof of the lowest whole-board energy use.

Available choices depend on driver and kernel configuration. Use scaling_available_governors rather than copying a name from another Pi.

Why is my Raspberry Pi CPU frequency low?

Use readings taken during the slowdown. Compare the firmware clock with the CPUFreq policy's minimum and maximum, and check current and historical flags separately.

Observation Check next
Low at idle, rises during work Normal scaling may explain it
Clock drops as temperature approaches the board's thermal limit during sustained work Temperature, airflow, and cooling
Current undervoltage flags Power supply and cable
Only historical flags remain Use the flag decoder and check whether current flags appear during the slowdown
Low policy maximum or powersave governor Check scaling_max_freq and scaling_governor before changing either
Normal clock, application still slow Memory, I/O, or application bottleneck

Do not diagnose thermal throttling from a historical bit alone. Do not compensate for insufficient power by forcing a higher frequency.

Test a governor reversibly

If you have a repeatable workload, time it under the current policy first. In Bash, record the original governor for policy0 and attempt a temporary change only if the new governor is listed:

1
2
3
4
5
6
7
policy_path=/sys/devices/system/cpu/cpufreq/policy0
candidate_governor=performance

previous_governor=$(cat "$policy_path/scaling_governor") &&
    printf 'Original governor: %s\n' "$previous_governor" &&
    grep -qw "$candidate_governor" "$policy_path/scaling_available_governors" &&
    printf '%s\n' "$candidate_governor" | sudo tee "$policy_path/scaling_governor"

If the change succeeds, run and time the same workload again. When it finishes, restore the original governor in that same shell and check the result:

1
2
3
test -n "$previous_governor" &&
    printf '%s\n' "$previous_governor" | sudo tee "$policy_path/scaling_governor" &&
    cat "$policy_path/scaling_governor"

Compare workload duration, clock, temperature and flags under both policies. For whole-board energy, measure power over the full workload. A service that manages CPU policy can overwrite a manual change.

Do I need to edit config.txt or rebuild the kernel?

Not for a first measurement or a temporary change to an available governor. Leave firmware clock and voltage settings at their defaults while diagnosing power and cooling.

Firmware configuration affects supported frequency behaviour but is not interchangeable with a Linux governor. Do not copy old arm_freq_min=100 advice across models. Kernel rebuilding is not a routine prerequisite for CPUFreq diagnosis.

Use the official hardware frequency-management documentation for model-specific controls.

Troubleshooting

Permission denied when writing the governor: the policy write needs administrative access. sudo echo ... > file does not elevate the shell's redirection; use the reviewed tee command above.

Invalid argument: confirm that the governor is listed for that policy. Do not force a name unavailable in the kernel.

No cpufreq directory: check on the Pi host, not only inside a container, and inspect its kernel/driver support. Installing a user-space utility cannot create a missing kernel policy.

Governor changes back: identify the service managing power policy before adding another persistent configuration. Two controllers can overwrite each other's settings.

Keep the default policy if a measured workload gains nothing from changing it.

Donate