nvitop

Overview

nvitop is an interactive NVIDIA device and process monitoring tool with a terminal interface that continuously updates device and process status. Beyond the monitor, the package provides Python APIs and a resource metric collector that reports per-GPU metrics including power usage in watts and milliwatts, plus a CUDA device selection tool (nvisel) for deep learning researchers. The project documents installation from PyPI, conda-forge or source, a GPU status snapshot, the metric collector and low-level device, process and host APIs. The library is released under Apache-2.0 while the command-line interface is GPL-3.0.

Editorial verdict

Fits researchers and platform engineers who want a terminal view of GPU processes plus a Python metric collector that records per-device power usage over time, with a CUDA device selection helper alongside. The trade-off is scope and licensing: it monitors NVIDIA GPUs only, produces no energy totals or carbon figures, and the split Apache-2.0 API and GPL-3.0 CLI means redistributing a bundled tool needs licence review. Power arrives in watts and milliwatts, so converting to energy and emissions stays in the consumer's own scripts. Teams needing fleet dashboards should export the metrics elsewhere.

Written by the GreenAIX editors from the recorded fields above. Vendors do not review or approve verdicts.

Use cases

• Watch GPU processes and utilization in a terminal
• Collect per-device GPU power usage metrics over time
• Build custom GPU monitoring scripts against the Python API
• Inspect GPU power limits and power status

Sources

• nvitop GitHub repository
• nvitop README
• nvitop documentation