Reading the table

All four resource columns cover the app’s whole process tree, not just the process Watchdog launched. They are the same figures Task Manager shows: CPU as a share of the whole machine, GPU as a share of one GPU (its busiest engine — 3D, video decode, copy), and VRAM as dedicated video memory on whichever GPU the app leans on hardest. GPU and VRAM read - on machines without GPU performance counters.

Under the table, a This PC strip shows what the whole machine is using. It displays the same four metrics across every running program — not just the supervised ones — turning amber past 75% and red past 90%.

It answers what the per-app columns cannot. In its own row alone, an app dropping frames because a backup job saturates the GPU looks identical to an app that is simply too heavy for the hardware. Hover any figure for detail, including which GPU it refers to.

The strip ends with TEMP, the GPU’s core temperature, separating a machine short of hardware from one short of airflow. A GPU running too hot keeps reporting ordinary load while its driver quietly cuts its clocks, dropping frames with nothing in the other numbers to explain it. When this happens, the reading says so outright — 93 °C (thermal).

The colour follows that card’s own limit rather than a fixed number, since cards start slowing down anywhere between about 85 and 100 °C. Hover it for the card’s name, the temperature at which it starts cutting clocks, the temperature at which it shuts itself off, and whether it is throttling right now.

Temperature is read from the NVIDIA driver. Other GPUs have no source for it that works without installing a kernel driver, so TEMP reads - on those machines. It also reads - while a display driver restarts, recovering on its own once the driver returns.