Skip to content

Power Management

Zion runs continuously, but most of the time it is lightly loaded. The goal is therefore simple: reduce idle power draw without suspending the host or making storage, networking, and guests unreliable.

What is optimized

  • CPU and package idle states; CPU frequency policy and Energy-Performance Preference (EPP);
  • PCIe, DMI and chipset link power management;
  • unused onboard devices;
  • standby behaviour of the mechanical data disk;
  • Linux power-saving tunables applied after boot.

Integrated graphics stays enabled for emergency console access and Intel Quick Sync. Zion itself is not suspended automatically.

Warning

This configuration was tested on Gigabyte Z370 HD3 with an Intel Core i5-8400. The same settings may behave differently on another board, controller, or kernel.

UEFI settings

CPU

Setting Value
Intel CPU C-States Control Enabled
Package C-State Limit C10
Intel Speed Shift Technology Enabled
Enhanced Halt State (C1E) Enabled
C3, C6/C7, C8 and C10 support Enabled
Enhanced Intel SpeedStep (EIST) Enabled
Race to Halt Enabled
Energy Efficient Turbo Enabled
Voltage Optimization Enabled

Prefetchers and ring-to-core offset settings remain on Auto. There was no workload-specific reason to change them.

Chipset and PCIe

Setting Value
PEG ASPM Enabled
PCH ASPM Enabled
DMI ASPM Enabled
CEC 2019 Ready Enabled
RC6 / Render Standby Enabled
Power Loading Auto

ASPM allows idle PCIe and chipset links to enter lower-power states. RC6 does the same for the integrated GPU when it is not processing video or driving a console.

Onboard devices

Device or feature Value
SATA Aggressive Link Power Management Enabled
Serial port Disabled
Parallel port Disabled
Onboard audio Disabled
Integrated graphics Enabled

Only confirmed unused devices are disabled.

ErP and Wake-on-LAN

ErP: Enabled reduces soft-off power consumption below the normal standby level, but disables Wake-on-LAN on this platform.

Zion is expected to remain online, so ErP is enabled. If remote Wake-on-LAN for Zion becomes necessary, ErP must be disabled and WoL tested again after a complete shutdown.

This does not affect Wake-on-LAN for the separate K3s nodes.

HDD spindown

The IronWolf disk stores backups and bulk data such as the media library. When no job or application is using it, stopping the platters saves several watts and allows the platform to reach deeper package states.

Persistent configuration should use the disk's /dev/disk/by-id/ path. Do not copy /dev/sdX names into configuration because they can change after a reboot.

Check the current state:

sudo hdparm -C /dev/disk/by-id/<IRONWOLF_DISK>

Request standby during a controlled test:

sudo hdparm -y /dev/disk/by-id/<IRONWOLF_DISK>

Set a ten-minute idle timeout:

sudo hdparm -S 120 /dev/disk/by-id/<IRONWOLF_DISK>

The disk wakes automatically on I/O. Backup jobs, SMART tests, media scans and filesystem activity can wake it as well.

Note

hdparm -S issued from the shell is not persistent. After adding the setting to the system configuration, verify it again after reboot.

Frequent spin-up and spin-down cycles defeat the purpose and add latency. If the disk wakes repeatedly during normal operation, increase the timeout or leave it running during active hours.

PowerTOP at boot

The tested PowerTOP tunables are reapplied by /etc/systemd/system/powertop.service:

[Unit]
Description=PowerTOP auto-tune and CPU power profile

[Service]
Type=oneshot
ExecStart=/usr/sbin/powertop --auto-tune
ExecStartPost=/bin/sh -ec 'for policy in /sys/devices/system/cpu/cpufreq/policy*; do echo powersave > "$$policy/scaling_governor"; echo balance_performance > "$$policy/energy_performance_preference"; done'
RemainAfterExit=true

[Install]
WantedBy=multi-user.target

Validate and enable the unit:

sudo systemd-analyze verify /etc/systemd/system/powertop.service
sudo systemctl daemon-reload
sudo systemctl enable --now powertop.service
sudo systemctl status powertop.service --no-pager

powertop --auto-tune changes several device policies at once. After a kernel update or hardware change, check the network interfaces, storage, USB devices and running guests. If one device becomes unstable, replace the blanket auto-tune with explicit settings for the tunables already verified as safe.

Validation

After changing the configuration, check:

  • CPU package C-state residency during a representative idle period;
  • network links and access to every VM and LXC;
  • NVMe and HDD access;
  • VM and container startup and shutdown;
  • backup execution and datastore availability;
  • HDD start/stop frequency;
  • failed systemd units;
  • reboot and recovery after power loss.

Useful read-only commands:

sudo powertop
sudo systemctl --failed --no-legend --plain --no-pager
sudo journalctl -b -u powertop.service --no-pager
sudo smartctl -a /dev/disk/by-id/<IRONWOLF_DISK>

Measured result

During the original tuning session:

  • CPU package idle states changed from no reported residency to regular pc2 and pc3 residency;
  • CPU C0 active time at idle fell from about 11% to about 3%;
  • idle draw of the motherboard and CPU platform fell by about 5.5 W, roughly 25% of that part of the baseline.

These figures describe one hardware and workload state. Measure again after adding PCIe devices, enabling passthrough, changing disks or moving substantial workloads between guests.

Wall power

Complete-system power was also measured at the wall:

System state Measured draw
Zion idle, with the IronWolf HDD spinning 22โ€“24 W
Zion under heavier workloads 40โ€“60 W
Dell Wyse 5070 idle, with two 2.5-inch drives 11โ€“12 W

Moving from the Dell Wyse 5070 to Zion roughly doubled idle consumption. In return, Zion provides a six-core CPU, NVMe storage, substantially more memory, hardware expansion and enough capacity to separate workloads into VMs and LXC containers. An idle draw of 24โ€“25 W is a reasonable trade-off for that increase in capability.

Operating rules

  • Stability and data integrity take priority over the lowest possible idle draw.
  • Backup windows take priority over HDD spindown.
  • UEFI changes are applied and tested in small groups.
  • ErP is reviewed whenever Wake-on-LAN requirements change.
  • PowerTOP is revalidated after kernel and hardware changes.
  • Measurements are compared with the same guests and storage state.