System Monitor

[system.monitor]
enabled = true                      # sample CPU, memory, network, load, temperature, GPU VRAM, and disk statistics
cpu_temp_sensor_path = ""           # exact temp*_input path for CPU temperature override; empty = auto
cpu_poll_seconds = 2.0              # CPU usage, CPU temperature, and load average
gpu_poll_seconds = 5.0              # GPU temperature, usage, and VRAM; probes run only while displayed
memory_poll_seconds = 2.0           # RAM usage
network_poll_seconds = 3.0          # network throughput
disk_poll_seconds = 10.0            # disk usage and swap usage

When disabled, Noctalia keeps system-monitor UI available but stops the background sampling thread. Bar sysmon widgets, desktop sysmon widgets, and the Control Center System tab show unavailable values until the setting is enabled again.

Poll intervals are configured globally under [system.monitor], not per widget. Each value is the number of seconds between samples for that metric group. A non-zero value is clamped to 1 to 120 seconds; setting a value to 0 disables that metric group entirely - it is never sampled and schedules no wakeups. GPU probes run only while UI displays a GPU stat, so the default interval does not wake an otherwise unused discrete GPU. Set gpu_poll_seconds = 0 to disable probing entirely. You can also adjust intervals in whole-second steps (down to 0) in Settings → Services → Advanced when the system monitor is enabled.

Key	Default	Metrics
cpu_poll_seconds	2.0	CPU usage %, CPU temperature, and load average (1 / 5 / 15 min)
gpu_poll_seconds	5.0	GPU temperature, usage %, and VRAM used/total. Probes run only while something displays a GPU stat; 0 disables them entirely
memory_poll_seconds	2.0	RAM usage % and used/total MB
network_poll_seconds	3.0	Network download and upload throughput
disk_poll_seconds	10.0	Disk used/free values for paths used by bar sysmon widgets, plus swap usage % and used/total MB

CPU usage and CPU temperature share cpu_poll_seconds because both are tied to the same CPU sampling pass. GPU temperature, usage, and VRAM share gpu_poll_seconds. Swap and disk usage share disk_poll_seconds. Network Rx and Tx share network_poll_seconds because they come from the same /proc/net/dev read. The aggregate network values exclude loopback; bar and desktop sysmon widgets can set interface to an exact /proc/net/dev interface name to show only that interface.

Bar and desktop sysmon widgets in graph mode, plus the Control Center System graphs, scroll using the same cadence as the fastest poll interval above (the minimum of the enabled, non-zero intervals). There is no separate graph-only setting. If every metric is set to 0, no history is recorded and the graphs stop updating.

CPU temperature auto-detection prefers known CPU hwmon drivers such as k10temp, zenpower, and coretemp. On AMD systems this normally selects the Tctl sensor exposed as temp1_input. To force a specific sensor, set cpu_temp_sensor_path to the exact sysfs input file:

[system.monitor]
cpu_temp_sensor_path = "/sys/class/hwmon/hwmon3/temp1_input"

When cpu_temp_sensor_path is set, Noctalia uses only that file. If the path is missing, unreadable, not a temp*_input file, or cannot be parsed, CPU temperature is reported as unavailable until the configuration is fixed.

To list available hwmon temperature inputs with their driver names, labels, and raw values:

for h in /sys/class/hwmon/hwmon*; do
  name="$(cat "$h/name" 2>/dev/null || echo unknown)"
  for input in "$h"/temp*_input; do
    [ -e "$input" ] || continue
    base="${input%_input}"
    label="$(cat "${base}_label" 2>/dev/null || basename "$base")"
    value="$(cat "$input" 2>/dev/null || echo unreadable)"
    printf '%s name=%s label=%s raw=%s\n' "$input" "$name" "$label" "$value"
  done
done

Available stats

Bar and desktop sysmon widgets and the Control Center System tab all draw from the same sampling thread, so they share one catalog of stats. Each widget picks one with stat (desktop graphs take an optional second with stat2).

Value	Display	Description	Bar	Desktop
cpu_usage	75%	CPU utilisation (all cores)	✓	✓
cpu_temp	65°C	CPU package temperature	✓	✓
gpu_temp	72°C	GPU temperature	✓	✓
gpu_usage	48%	GPU utilisation	✓	✓
gpu_vram	42%	GPU VRAM usage	✓	✓
ram_used	4.2 GiB	RAM used (binary, auto-scaled)	✓	—
ram_pct	26%	RAM used %	✓	✓
swap_pct	12%	Swap used %	✓	✓
disk_used_pct	45%	Disk used % for the widget’s path	✓	—
disk_used	120 GiB	Disk space used for the widget’s path	✓	—
disk_free_pct	55%	Disk free % for the widget’s path	✓	—
disk_free	380 GiB	Disk space available for the widget’s path	✓	—
net_rx	1.2 MB/s	Network download speed (decimal, auto-scaled)	✓	✓
net_tx	256 kB/s	Network upload speed (decimal, auto-scaled)	✓	✓

A stat whose source is unavailable on the machine - GPU VRAM on an integrated GPU, for example - is shown as unavailable rather than as zero. What each GPU vendor can report is covered under GPU data sources.

The stats are read once and shared, but a few presentation options stay per widget, because two widgets showing the same stat may want to show it differently: path selects which mount the four disk stats measure, and interface selects which kernel interface net_rx/net_tx measures (an exact name from /proc/net/dev such as wlan0 or enp5s0; empty means the total across all non-loopback interfaces). Both are documented with the widget settings.

GPU data sources

Noctalia reads GPU statistics straight from the kernel and from the vendor’s own management library where one exists. No elevated privileges are required, and Noctalia does not invoke command-line monitoring tools such as nvidia-smi, intel_gpu_top, or radeontop.

Integrated GPUs are supported

GPU monitoring is not limited to discrete GPUs. Noctalia can report temperature and usage for supported integrated AMD and Intel GPUs when their kernel drivers expose those metrics. The gpu_vram stat is different: an integrated GPU that shares system memory has no dedicated VRAM to report, so that one stat is unavailable even though its other GPU stats can work.

GPU	Runtime requirement	Extra package needed?
AMD	The kernel amdgpu driver and its sysfs interfaces	No. The optional ROCm SMI library (librocm_smi64.so) is preferred when installed, but the normal kernel interfaces are sufficient
NVIDIA	The proprietary NVIDIA driver and NVML (libnvidia-ml.so.1)	Usually supplied by the driver’s userspace utilities package. On Arch Linux, install nvidia-utils
Intel	The kernel i915 or xe driver and DRM interfaces	No

For NVIDIA, the driver and userspace library must be compatible; installing nvidia-utils alone does not replace the matching kernel driver. You can check that the required NVML library is visible to applications with:

ldconfig -p | grep 'libnvidia-ml.so.1'

The open-source Nouveau driver does not provide NVML, so NVIDIA usage and VRAM monitoring require the proprietary driver. The nvidia-smi executable may be installed by the same package, but Noctalia loads NVML directly and does not run it.

GPU	Temperature	Usage	VRAM
AMD	amdgpu hwmon	gpu_busy_percent	mem_info_vram_used / mem_info_vram_total
NVIDIA	NVML, hwmon otherwise	NVML	NVML
Intel	i915 / xe hwmon	per-client DRM statistics	xe device query

On AMD, ROCm SMI is used instead of sysfs when the ROCm runtime is installed. On a hybrid laptop whose discrete NVIDIA GPU is asleep, NVML is skipped so the GPU is not woken, and the integrated GPU is reported instead.

Intel is the odd one out: it publishes no device-wide busy counter, so usage is summed from the per-client statistics the kernel exposes for each process using the GPU. Three consequences follow. Work done by other users’ processes is not counted, which is invisible on a normal single-user desktop. Usage needs two samples before it can report anything, so the first value appears one gpu_poll_seconds interval after startup. Noctalia currently counts Intel render and compute engines, but not media/video engines; hardware-accelerated video playback can therefore show 0% while a per-engine monitor that includes video decoding reports activity. This difference does not mean the integrated GPU is unsupported.

Intel VRAM is reported only for discrete cards such as Intel Arc, and requires Linux 6.18 or newer — earlier kernels release the used-memory figure only to privileged processes, so Noctalia reports VRAM as unavailable rather than showing a permanent zero. Integrated Intel GPUs have no dedicated video memory and never report VRAM.

Value highlighting

Bar sysmon widgets tint their value (icon, text label, graph line, and gauge fill) toward the widget’s highlight_color as the stat rises. Each stat has an activity threshold and a critical threshold, configured globally under [system.monitor]:

[system.monitor]
cpu_usage_activity_threshold = 50     # %
cpu_usage_critical_threshold = 90     # %
cpu_temp_activity_threshold  = 60     # °C
cpu_temp_critical_threshold  = 85     # °C
gpu_temp_activity_threshold  = 60     # °C
gpu_temp_critical_threshold  = 85     # °C
gpu_usage_activity_threshold = 50     # %
gpu_usage_critical_threshold = 95     # %
gpu_vram_activity_threshold  = 50     # %
gpu_vram_critical_threshold  = 90     # %
ram_pct_activity_threshold   = 60     # %
ram_pct_critical_threshold   = 90     # %
swap_pct_activity_threshold  = 20     # %
swap_pct_critical_threshold  = 80     # %
disk_used_pct_activity_threshold = 80 # %
disk_used_pct_critical_threshold = 95 # %
disk_used_activity_threshold     = 80 # %
disk_used_critical_threshold     = 95 # %
disk_free_pct_activity_threshold = 80 # %
disk_free_pct_critical_threshold = 95 # %
disk_free_activity_threshold     = 80 # %
disk_free_critical_threshold     = 95 # %
net_rx_activity_threshold    = 1      # MB/s
net_rx_critical_threshold    = 50     # MB/s
net_tx_activity_threshold    = 1      # MB/s
net_tx_critical_threshold    = 50     # MB/s

The value’s color is interpolated between the widget’s regular foreground color (its color setting) and its highlight_color (set per widget - see the sysmon widget). Below the activity threshold the value stays at its regular color; the moment it crosses the activity threshold it jumps to a small highlight tint, then ramps the rest of the way to the full highlight_color, reaching it at or above the critical threshold. Set the critical threshold to 0 to disable highlighting for that stat.

Key	Default	Unit	Stat
cpu_usage_activity_threshold / cpu_usage_critical_threshold	50 / 90	%	CPU usage
cpu_temp_activity_threshold / cpu_temp_critical_threshold	60 / 85	°C	CPU temperature
gpu_temp_activity_threshold / gpu_temp_critical_threshold	60 / 85	°C	GPU temperature
gpu_usage_activity_threshold / gpu_usage_critical_threshold	50 / 95	%	GPU usage
gpu_vram_activity_threshold / gpu_vram_critical_threshold	50 / 90	%	GPU VRAM usage
ram_pct_activity_threshold / ram_pct_critical_threshold	60 / 90	%	RAM usage (also ram_used)
swap_pct_activity_threshold / swap_pct_critical_threshold	20 / 80	%	Swap usage
disk_used_pct_activity_threshold / disk_used_pct_critical_threshold	80 / 95	%	Disk used %
disk_used_activity_threshold / disk_used_critical_threshold	80 / 95	%	Disk used
disk_free_pct_activity_threshold / disk_free_pct_critical_threshold	80 / 95	%	Disk free %
disk_free_activity_threshold / disk_free_critical_threshold	80 / 95	%	Disk free
net_rx_activity_threshold / net_rx_critical_threshold	1 / 50	MB/s	Network download
net_tx_activity_threshold / net_tx_critical_threshold	1 / 50	MB/s	Network upload

The ram_used and ram_pct stats share the ram_pct_* thresholds, since both are driven by the same RAM usage percentage. You can also adjust every threshold in Settings → Services when the system monitor is enabled; each stat has a single dual-handle range slider - the left handle sets the activity threshold and the right handle the critical threshold, and the handles cannot cross, so the activity threshold can never exceed its critical threshold.