Nsight Systems'ta Zaman Çizelgesi Görünümü

Nsight Systems is your bird's-eye profiler. It shows the whole timeline of your program so you can spot where the GPU sits idle. 🔭

A timeline lays every operation along a time axis. You instantly see what runs when, and how long each piece takes, side by side.

You record a run with nsys on the command line. It captures GPU and CPU activity into one report you can open later.

Each row is a stream of activity: CPU threads, CUDA API calls, kernels, and memory copies. Reading them together tells the story.

Empty space on the GPU row is a gap: time the GPU did nothing. Gaps are your biggest clue that something is waiting. ⏳

On the timeline, memory copies and kernels live on separate lanes. If copies dominate, your data movement is the real cost.

When copies and kernels run at the same time, their bars overlap on different streams. Overlap is good: the GPU is staying busy.

The big picture hides detail, so you zoom in on a slow region. Up close, you can measure one kernel or one copy precisely.

Hovering any bar shows its exact duration and name. That turns a vague feeling of slowness into a hard number you can act on.

Sometimes the GPU waits because the CPU is busy preparing work. The timeline reveals these host-side stalls just as clearly.

Nsight Systems answers where the time goes across the whole app. For deep per-kernel detail you switch to a different tool.

You see a long empty stretch on the GPU row.

Nsight Systems gives you the timeline: gaps, copies, kernels, and overlap at a glance. Start here to find where your time really goes. 👏