Intel GPU Monitor

A dank bar widget that monitors an Intel GPU (integrated or discrete Arc). It shows GPU usage, temperature and (optionally) VRAM with your choice of chart, and opens a detail view listing the processes using the GPU — similar in spirit to System Monitor Plus.

No setup required

Everything is read from the kernel — no installation, no root, no special permissions, and reading the counters does not wake the GPU:

• GPU usage % and the per-process list come from DRM fdinfo — the per-process GPU engine counters (drm-engine-*) that intel_gpu_top/nvtop use. A single awk pass over /proc/*/fdinfo/*, filtered to the Intel driver (i915/xe) so a discrete NVIDIA/AMD GPU is never counted, gives the busiest engine's real busy % (render, blitter, video, video-enhance, compute).
• VRAM comes from the same scan (drm-resident-*): shared system memory on iGPUs, dedicated (local) memory on discrete Arc.
• Temperature comes from the GPU's hwmon sensor if present (discrete cards), else the CPU package/die temperature (integrated GPUs have no separate sensor — shown as GPU Temp (die)). Frequency comes from sysfs. Both are read with Quickshell FileView (no process spawned).

intel_gpu_top is optional — only used for the "open in terminal" action, if you enable it. It is not needed for the widget.

Installation

git clone https://github.com/rdannenbring/dms-intel-gpu-plugin ~/.config/DankMaterialShell/plugins/intelGpuMonitor

Then enable Intel GPU Monitor in DMS Settings → Plugins and add it to a bar.

Features

• GPU usage % in the bar (toggle), with a chart (toggle) — chart type: bar, gauge, donut, pie or thermometer. Charts are GPU-drawn (QtQuick.Shapes / rectangles, no CPU Canvas); bar is the lightest.
• Temperature value in the bar (toggle), with a chart (toggle) — same chart types, configurable °C range.
• VRAM % in the bar (toggle) — off by default; see the note below.
• In-chart label (toggle, per metric): show an icon (custom picker) or a letter inside/above each chart to identify it — centered in gauge/donut, above the element on bars/thermometer.
• Configurable mouse actions (left / right / middle), each: detail view, menu, open in terminal, or nothing. Defaults: left = detail, right = menu, middle = nothing.
• Open intel_gpu_top in a terminal — defaults to your $TERMINAL; the run flag is chosen automatically for known terminals.
• Detail view: large charts for the enabled metrics, a per-engine breakdown, and a table of processes using the GPU — Process, GPU %, VRAM (MB), VRAM %, PID.

Performance

The metrics refresh on an interval (Refresh interval, default 2 s). Each refresh is one find | awk scan of the DRM fdinfo files under /proc (~200 ms of mostly disk-cache I/O, off the render thread) plus a couple of tiny sysfs reads. The file list is piped from find (never passed as shell arguments), so the scan is unaffected by how many file descriptors are open system-wide. It pauses while the bar is hidden. If your GPU has little headroom, raise the interval (e.g. 5000 ms) and prefer the bar chart type or numbers only.

VRAM notes

• VRAM comes from the same fdinfo scan as usage — no extra cost.
• Integrated GPUs report shared system memory, so the percentage is of system RAM and reads low; the used-MB figure is the real GPU-resident memory.
• Discrete Intel Arc GPUs report dedicated (local) memory. Set your card's VRAM in Total VRAM override (MB) (e.g. 8192 for an 8 GB A770) so the percentage is meaningful. ⚠️ Discrete-Arc handling is implemented but untested — the developer has no Arc card. Feedback welcome.

Settings overview

• General: refresh interval.
• GPU Usage / VRAM / Temperature (each on its own tab): show value, show chart (+ chart type & bar width), in-chart label (icon or letter), plus per-metric extras (temperature source GPU & range, VRAM override).
• Interaction: left / right / middle click actions, terminal launcher toggle and terminal command.

Backlog

• Per-value thresholds to drive icon changes, color changes and alert messages.

License

MIT — see LICENSE.