Skip to content

Hardware

A click on the host card at the top of the sidebar (hostname, OS, uptime) opens Hardware: a spec sheet of the machine, read only. It is also in the command palette (Ctrl+K → Hardware). The Power tab shows what the server draws and what that costs (see below).

On one screen

  • At a glance: CPU, memory, GPUs, mainboard and BIOS.
  • Processor: model, cores and threads, maximum clock, L3 cache, and whether hardware virtualisation (VT-x/AMD-V) is available for virtual machines.
  • Memory: the slots as boxes – filled with size, type, speed and module, or free – plus the maximum the board supports and whether ECC is active. Useful to plan an upgrade without opening the case. Slots come from dmidecode (root, in the helper); without it only the total is shown.
  • Mainboard and BIOS: vendor, model, form factor, BIOS version and date. A BIOS that is four years or older gets a hint to look for an update.
  • Graphics: every GPU with driver and device nodes, and the ready-made Quadlet line (AddDevice=/dev/dri/renderD128) that gives a container such as Jellyfin or Immich the GPU for hardware transcoding, with a copy button.
  • Sensors: temperatures (with a bar up to the critical value), fan speeds ("stopped" for a stopped fan), labelled voltages and power, grouped by chip, refreshed every 15 seconds.
  • USB devices: name, maker, USB version, driver – and for serial devices (Zigbee and Z-Wave sticks) the stable path under /dev/serial/by-id/… with a copy button. That path survives reboots, unlike ttyUSB0, and belongs into Home Assistant, Zigbee2MQTT or AddDevice=.
  • SATA links: per disk the negotiated speed. 3 Gbit/s on a 6 Gbit/s disk and port usually means a cable or port problem – the same cause as rising CRC errors on the disks page.
  • PCIe devices: grouped (NVMe, storage controllers, network, graphics, …) with the kernel name (nvme0, enp5s0), driver and link (x4 · PCIe 4.0).

Warnings

  • An NVMe drive, storage controller or network card running with fewer lanes than it supports (runs at x1 instead of x2) or an NVMe drive on an older PCIe generation – usually a slot with fewer lanes or lanes shared with another slot (see the mainboard manual). GPUs are left out: they lower their link speed when idle.
  • A SATA link slower than disk and port allow.
  • A temperature within 5 °C of its critical value.
  • An old BIOS (hint), and a note when Quadeck runs in a virtual machine, where all of this shows the virtual hardware.

Where the data comes from

Everything is read from sysfs and procfs – /sys/bus/pci, /sys/bus/usb, /sys/class/drm, /sys/class/ata_link, /sys/class/hwmon, /sys/class/dmi/id, /proc/cpuinfo, lscpu – so lspci and lsusb are not needed. Device names come from pci.ids/usb.ids (package hwdata) when installed, otherwise a built-in list names the usual vendors. Missing sensors usually mean a missing kernel module for the board's sensor chip (nct6775, it87).

Power

Hardware → Power: what the server draws now, per component, and what it used over time. There is no meter at the wall socket, so the total is always marked estimated; each part says whether it is measured or estimated.

  • CPU – measured from the processor's own energy counter (RAPL, /sys/class/powercap/intel-rapl:*, Intel and AMD). Packages plus DRAM where Intel reports it apart; never psys (the whole platform) or the core/uncore sub-zones, which are inside the package. The counters are root-only, so the helper reads them every 30 s; watts are the difference over that time, so no short spike is missed. Without readable counters (VMs, ARM boards) the CPU is estimated from its load (4 W idle up to about 34 W) and marked so.
  • GPU – measured as the driver reports it (nvidia-smi power.draw, amdgpu/i915 hwmon), the same value as on the GPU card. Integrated graphics without a power reading count as 0 W; an AMD APU's reading may include part of the CPU.
  • Disks – estimated from type and state, because disks have no power sensor: a spinning hard disk 6 W, in standby 0.8 W (the state comes from hdparm -C, which does not wake the disk; disks behind USB are not asked every 30 s – some bridges wake them – and count as spinning), a SATA SSD 1.2 W, an NVMe 3 W.
  • Rest – estimated: a base value for mainboard, RAM, fans and network (15 W by default) plus the power supply loss (10 % of everything by default).

Below that:

  • Today, this month (with a projection to the month's end) and per year, in kWh and euros from the electricity price you enter.
  • Usage as stacked bars per component: 24 hours by the hour, 7 or 30 days by the day, 12 months by the month, with the peak.
  • Disks in detail: type, state, estimated watts and standby hours today per disk, and how much disk standby saves a month (standby hours × the difference between spinning and standby), with a link to the disks' energy saving.

Settings …: the electricity price (€/kWh), the base value (W) and the power supply loss (%). If you once measured the server with a meter at the socket, set the base value so that Now matches.

The overview gets a Power card (watts, kWh per day, € per month) once the first reading is in; like the other cards it keeps a 7-day chart, its dialog links here.

Energy is stored as Wh per hour and component (plus standby seconds per disk) in SQLite and kept for two years; the current hour is rewritten every 30 s, so a restart loses nothing. Time when the server or Quadeck was off is not counted.