DGXTOP for The DGX SPARK

Fork Notice: This is a fork of GigCoder-ai/dgxtop with enhanced GPU process monitoring features. Maintained by SonusFlow AI.

A performance monitoring CLI tool for Ubuntu inspired by asitop for Mac, with added volume transfer speed monitoring capabilities.

Features

• GPU Process Monitoring: View running GPU processes with PID, user, GPU/CPU usage, memory (nvtop-style)
• Interactive Process Management: Navigate, sort, and kill GPU processes
• Volume Transfer Speed Monitoring: Real-time read/write speed tracking per drive
• System Monitoring: GB10 GPU, CPU, memory, and network statistics
• Real-time Display: Interactive terminal interface with customizable update intervals
• Lightweight: Minimal dependencies, uses native Linux /proc filesystem
• Per-Drive Performance: Detailed breakdown of I/O performance for each storage device

Installation on DGX Spark

Prerequisites

• NVIDIA DGX Spark with Ubuntu
• Python 3.8+
• nvidia-smi (included with NVIDIA drivers)

Quick Install (Build from Source)

SSH into your DGX Spark and run:

# 1. Install build dependencies
sudo apt update
sudo apt install -y git debhelper dh-python python3-all python3-setuptools dpkg-dev

# 2. Clone the repository
git clone https://github.com/sonusflow/dgxtop.git
cd dgxtop

# 3. Build the .deb package
dpkg-buildpackage -us -uc -b

# 4. Install
sudo apt install ../dgxtop_1.0.0-1_all.deb

Verify Installation

dgxtop --version
dgxtop

Uninstall

sudo apt remove dgxtop

Upgrade

cd dgxtop
git pull
dpkg-buildpackage -us -uc -b
sudo apt install --reinstall ../dgxtop_1.0.0-1_all.deb

Original Repository

For the original version without fork enhancements, see GigCoder-ai/dgxtop

Usage

Basic Usage

dgxtop

Options

dgxtop --interval 0.5        # Update every 0.5 seconds
dgxtop -i 2.0                # Update every 2 seconds
dgxtop --version             # Show version information

Interactive Controls

Key	Action
j / k	Navigate up/down in process list
s	Enter sort mode (cycle sort columns)
K (Shift+k)	Enter kill mode for selected process
Enter	Confirm action
Esc	Cancel mode
+ / -	Speed up / slow down refresh
q	Quit

Architecture

dgxtop/
├── __init__.py              # Package initialization
├── main.py                  # Main application entry point
├── gpu_monitor.py           # GPU monitoring (nvidia-smi)
├── gpu_processes_monitor.py # GPU process monitoring (nvtop-style)
├── disk_monitor.py          # Disk I/O monitoring (/proc/diskstats)
├── system_monitor.py        # CPU, memory, network monitoring
├── network_monitor.py       # Network interface monitoring
├── rich_ui.py               # Terminal UI (Rich library)
├── config.py                # Configuration management
└── logger.py                # Logging system

Fork Enhancements

This fork adds the following features not in the original:

• GPU Process Panel: Displays running GPU processes with:
  • PID, User, Device, Type (Compute/Graphic)
  • GPU utilization %, GPU memory usage
  • CPU %, Host memory usage
  • Full command line
• Interactive Controls: vim-style navigation (j/k), sort mode, kill mode
• Visual Selection: Highlighted process selection with reverse video

Technical Details

Volume Transfer Speed Calculation

The tool calculates transfer speeds by reading /proc/diskstats at regular intervals:

Read/Write Bytes per Second = (Δsectors) × 512 bytes / Δtime

Data Sources

• Disk Statistics: /proc/diskstats
• CPU Statistics: /proc/stat
• Memory Statistics: /proc/meminfo
• Network Statistics: /proc/net/dev
• GPU Statistics: nvidia-smi
• GPU Processes: nvidia-smi --query-compute-apps and nvidia-smi pmon

Requirements

• Python 3.8+
• DGX Spark with NVIDIA Ubuntu
• Rich library (installed automatically)

License

Apache 2.0 License - see LICENSE file for details.

Credits

• Original Author: GigCoder-ai
• Fork Maintainer: SonusFlow AI
• Inspired by asitop for Mac

Roadmap

• Add GPU process monitoring
• Add interactive process navigation
• Add sort and kill functionality
• Implement alerting thresholds
• Add configuration file support
• Create systemd service option