dsh-gpu-pulse

When you run long jobs inside the DSH Web UI, GPU utilization, VRAM pressure, and temperature are the first things you want to glance at without leaving the page. dsh-gpu-pulse mounts a compact, floating monitor strip directly onto the DSH page, pulling data from nvidia-smi on the DSH host via a lightweight poll-and-cache route—no extra daemon, no agent calls, no build step.

One row per GPU lays out utilization, used/total VRAM, temperature, power draw, and the top VRAM-consuming processes when the driver reports them, each colour-coded green, amber, or red against sensible thresholds. The whole strip is press-and-drag anywhere on the page; the position is persisted per browser profile and a double-click snaps it back to the default corner. It always renders above side panels and drawers, so it never gets buried.

This is built for anyone running NVIDIA discrete GPUs on Windows or Linux who works with DSH daily—especially multi-GPU inference or training setups where you want all cards visible at a glance. On machines without an NVIDIA driver it degrades to a small n/a pill and quietly re-probes every five minutes, lighting up on its own the moment a driver appears, with zero manual follow-up.