igputop: htop-style Intel iGPU monitoring on Lunar Lake (xe)

Why

On Lunar Lake the GPU runs the **xe driver. A lot of older tooling assumed `i915/i915_pmu`, so you can end up with no useful live view** of what is using the iGPU.

The kernel still exposes the truth: **drm-* lines in `/proc/<pid>/fdinfo** plus the client list under/sys/kernel/debug/dri/.../clients`. **igputop reads those directly: per-process memory, engine activity (render / video / …), two GT frequencies (gt0, gt1), and power when a counter actually moves (with a clear package proxy** fallback if uncore is stuck).

How

1. Clients — who has the GPU open.
2. *fdinfo `drm-** — memory buckets + engine counters (xe` uses cycle counters → utilization is delta ratios between refreshes).
3. Dedup — do not sum every fd; one client id can appear on many fds. **drm-client-id** collapses that.
4. Lunar Lake — two GTs, separate freqs; RAPL uncore sometimes does not tick → optional package proxy, labeled.

UI

Top: card, PCI pdev, driver, memory, freq, power. Middle: engine bars. Bottom: process table (default sort: GTT memory).

Keys: q quit · s sort · r refresh · 1–9 pick card. **--batch** = one text snapshot (SSH-friendly).

Install (binary only)

Needs root or **CAP_SYS_ADMIN** (debugfs clients file).

curl -fL -o igputop-linux-amd64 'https://github.com/ruhailrafiq7/ExporterProject/raw/main/igputop-linux-amd64'
chmod +x igputop-linux-amd64
sudo install -m 0755 igputop-linux-amd64 /usr/local/bin/igputop
sudo igputop                    # or: sudo setcap cap_sys_admin+ep /usr/local/bin/igputop && igputop

If empty: sudo mount -t debugfs none /sys/kernel/debug then sudo ls /sys/kernel/debug/dri/*/clients.

Caveats

Engine % needs two samples (first tick may show —). Power may read package (proxy). Linux only.