nvitop

nvitop is a terminal GPU monitor for people running AI workloads on NVIDIA hardware. It shows GPU activity alongside the processes using it, so you can check resource use and investigate busy devices.

• Live monitoring: View GPU utilization, memory use and host process metrics with bar charts and history graphs, then sort or filter processes to focus on a workload.
• Process inspection and control: Inspect environment variables and a tree of GPU processes, their children and ancestors, or send signals directly from the monitor.
• Custom metrics and dashboards: Python APIs collect host, GPU and process metrics for your own monitoring tools, while nvitop-exporter connects monitoring to Prometheus and Grafana dashboards.
• GPU selection: The included nvisel tool chooses GPUs against supplied criteria and can pass that selection to a training script.

You can use an interactive view with keyboard and mouse controls or get a single status snapshot. The process tree covers GPU processes and their relatives rather than every process on the machine.

nvitop runs locally on Linux and Windows, works over SSH, and supports Docker with NVIDIA container support. It's open source under Apache 2.0 and requires NVIDIA hardware compatible with NVIDIA Management Library (NVML), plus access to that library through NVIDIA drivers or the CUDA Toolkit. Mouse support and other terminal behavior can vary between Windows terminal emulators.