nvsonar 1.0.2

Active GPU diagnostic tool that identifies performance bottlenecks

NVSonar

Active GPU diagnostic tool that identifies performance bottlenecks using targeted micro-probes.

Why NVSonar?

Traditional GPU monitoring tools show utilization percentages, but this can be misleading. A GPU reporting 100% utilization may actually be computing useful work, or it may be stalled waiting on memory transfers (memory-bound) or PCIe transfers (PCIe-bound).

Features

• Runs CUDA micro-probes to stress-test specific GPU subsystems
• Monitor temperature, power, utilization, clocks
• Provides an interactive TUI with multi-GPU support
• Detects performance bottlenecks in memory, compute, or PCIe
• Generates an overall GPU health and performance score

Installation

pip install nvsonar

Quick Start

# Launch interactive TUI with all GPUs and live metrics
nvsonar

Interface

┌─ NVSonar ──────────────────────────────────────────────────────┐
│                          Available GPUs                        │  
┡━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━╇━━━━━━━━━━━╇━━━━━━┩
│ Index │ Name                       │ Memory │ Driver    │ CUDA │
│   0   │ NVIDIA GeForce GTX 1650 Ti │ 4.0 GB │ 580.95.05 │ 13.0 │
└───────┴────────────────────────────┴────────┴───────────┴──────┘

╭─────────────────────── GPU 0 Metrics ──────────────────────────╮
│  Temperature         45.0°C                                    │
│  Power               2.8W                                      │
│  GPU Utilization     0%                                        │
│  Memory Utilization  0%                                        │
│  Memory Used         0.4 / 4.0 GB                              │
│  GPU Clock           300 MHz                                   │
│  Memory Clock        405 MHz                                   │
╰────────────────────────────────────────────────────────────────╯

• All available GPUs displayed in table at top
• Live metrics for each GPU (updates every 0.5s)
• Press 'q' to quit

Requirements

• Python 3.10+
• NVIDIA GPU with driver installed
• CUDA toolkit (for active probes)
• Linux (tested on Ubuntu)

License

Apache License 2.0 - see LICENSE for details.