CPU Power Mode

Astra sends packets on a strict schedule. When the CPU runs below its top speed, packets go out late and the output jitters under load. On servers, set the CPU to performance mode.

Dashboard warning

The web dashboard shows this alert when the CPU power policy of a host holds its clock down:

The power saving policy holds the CPU on <host> below the clock it is capable of. This could impact application performance

What triggers it depends on the CPU driver:

• intel_pstate and amd-pstate-epp: the alert appears only when energy_performance_preference is power or balance_power. The powersave governor is fine here: the hardware picks the frequency and turbo still works.
• Other drivers (acpi-cpufreq, cppc_cpufreq, intel_cpufreq): the alert appears for every governor except performance. Load-following governors (ondemand, schedutil, conservative) keep a bursty streaming load at the lowest clock and ramp up only under sustained load.

Disabled turbo does not raise the alert. To fix it, switch to performance mode. “Read more” in the alert opens the Tune page.

“Do not show again” hides the alert for your user on this host. To see it again, clear “Hide CPU power save warning” in Settings → Interface.

Check the current mode

The cpupower utility prints the driver, the available governors, and the current policy. On Debian it comes in the linux-cpupower package:

sudo apt-get update
sudo apt-get install linux-cpupower

On Ubuntu it is in linux-tools-common plus the package for the running kernel:

sudo apt-get update
sudo apt-get install linux-tools-common linux-tools-$(uname -r)

cpupower frequency-info

You can also read these values directly from sysfs. Checking one CPU is enough - in practice the governor is the same for all CPUs on a machine:

cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_driver
cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor
cat /sys/devices/system/cpu/cpu0/cpufreq/energy_performance_preference

Driver	Value that matters	Fine when
intel_pstate, amd-pstate-epp	energy_performance_preference	default, performance or balance_performance
acpi-cpufreq, cppc_cpufreq, intel_cpufreq, others	scaling_governor	performance
no cpufreq directory (most VMs)	nothing to check	the guest cannot set the clock

Note

The cpu MHz value in /proc/cpuinfo doesn’t prove anything on an idle CPU - every governor lets an idle core drop to a low clock. It only shows the real policy under load.

Switch to performance

Note

You can use our Tune script to automatically configure these settings. Read more

Tune sets every scaling_governor to performance and applies this at every boot. This is the recommended fix for servers. To do the same by hand:

cpupower frequency-set -g performance

On intel_pstate and amd-pstate-epp this also sets energy_performance_preference to performance.

On a desktop or laptop, the desktop environment usually controls this preference: power-profiles-daemon (behind the “Power Mode” setting in GNOME and KDE), or TLP. The Power Saver profile sets energy_performance_preference to power, and TLP also lowers it on battery power. Switch the profile to performance or balanced:

powerprofilesctl set performance

The preference can also be set directly while keeping the governor:

echo balance_performance | sudo tee /sys/devices/system/cpu/cpu*/cpufreq/energy_performance_preference

Valid values are default, performance, balance_performance, balance_power, and power.

Settings made with cpupower or tee don’t survive a reboot. To make them permanent, add the command to /etc/rc.local so it runs at every boot. Tune already does this for the governor automatically.