nvitop is a command-line tool that provides an interactive viewer for monitoring NVIDIA GPU processes and resource utilization.
The tool addresses the limitation of nvidia-smi by offering a more detailed, interactive interface for tracking GPU processes in real time. It uses the NVIDIA Management Library to query GPU state and presents this information through a curses-based terminal interface similar to the Unix top utility, allowing users to monitor which processes are consuming GPU memory and compute resources, kill processes, and inspect detailed GPU metrics without leaving the terminal.
Teams managing GPU-intensive workloads should consider nvitop if they need better visibility into per-process GPU usage than nvidia-smi provides. It suits environments where developers or operators regularly need to diagnose GPU resource contention, identify runaway processes, or monitor training jobs. The tool also includes nvitop-exporter, which exposes GPU metrics in Prometheus format and integrates with Grafana dashboards for centralized monitoring across multiple machines.
The project maintains active development with regular updates addressing user-reported issues and feature requests. The codebase shows consistent refinement of the monitoring interface and expansion of export capabilities. Documentation is comprehensive and kept current alongside code changes. The maintainer engages responsively with the community through issue tracking and incorporates feedback into releases.
