XuehaiPan/nvitop
Nvitop
Features
- NVML-Based GPU Abstractions - Wraps NVML calls into Python objects that cache GPU device and process properties for efficient access.
- GPU Process Managers - Ships an interactive terminal interface for sorting, filtering, and terminating GPU processes with keyboard shortcuts.
- GPU Monitoring APIs - Provides a Python API to query GPU device properties, process details, and host metrics.
- Terminal Interaction - Provides a real-time terminal interface with keyboard shortcuts for managing GPU processes and devices.
- GPU Device Metric Queries - Reads per-device GPU statistics like fan speed, temperature, utilization, and memory usage via a Python API.
- Metric Collection Daemons - Provides a daemon thread that periodically collects GPU and host metrics and delivers them to a callback.
- GPU Resource Monitoring - Provides real-time monitoring of GPU utilization, memory, temperature, and power metrics through a terminal interface.
- Real-Time Monitors - Displays live resource usage for all NVIDIA GPUs with continuous updates.
- GPU Process Monitors - Displays live GPU process information including memory usage and compute utilization for each device.
- Background Resource Metric Collectors - Gathers host and GPU resource metrics periodically on a background thread and delivers them to a callback.
- Interactive Process Managers - Provides an interactive terminal interface to sort, filter, tag, and signal GPU processes.
- GPU Device and Process Snapshots - Retrieves a combined snapshot of GPU device status and running processes in a single call.
- GPU Device Property Queries - Reads static and dynamic attributes of GPU devices such as driver version, CUDA version, and memory capacity.
- Interactive GPU Resource Managers - Kills processes or adjusts GPU settings interactively from the terminal.
- Programmatic GPU Query APIs - Provides a Python API for querying GPU device properties, process details, and resource metrics programmatically.
- Metric Collection - Runs a background daemon thread that periodically collects resource metrics.
- Device Snapshots - Retrieves structured snapshots of GPU device and process status for analysis.
- Process State Inspection - Represents OS processes by PID, caching instances and providing access to metadata for GPU process management.
- Prometheus-Based Metric Exporters - Exposes GPU and host metrics as Prometheus-format endpoints for integration with monitoring stacks.
- GPU Process Analysis - Exposes GPU-specific resource usage and metadata for individual running processes.
- Colorized GPU Status Outputs - Returns detailed, colorized information about GPU devices and their processes programmatically.
- GPU Process Management Interfaces - Lists running GPU processes with PID, user, and memory, and allows termination from the interface.
- System Monitors - Interactive NVIDIA GPU process viewer.
