dubeyakshat07 / gpu-hardware-profiling-and-monitoring

GPU Hardware Profiling and Monitoring Tool

It provides a comprehensive, cross-platform toolkit to capture GPU hardware details, profile the software stack, and implement active, real-time monitoring with a web-based dashboard.

It is designed to work on both Linux, MacOS and Windows and can detect GPUs from major vendors like NVIDIA, AMD, Intel, and Google TPUs (on Linux).

Features

• Platform Profiling: Gathers detailed information about:
  • GPU Hardware: Vendor, model, memory (supports NVIDIA, AMD, Intel, and Google TPUs).
  • System: OS, architecture, hostname.
  • Software Stack: GPU drivers, CUDA version, OpenCL availability.
• Workload Monitoring: Provides real-time metrics for:
  • GPU utilization, memory usage, temperature, and power draw.
  • System-level stats like CPU and RAM usage.
  • A list of processes currently using the GPU (NVIDIA only).
• Web Dashboard: A clean and modern web interface to visualize all monitoring data in real time.
• Alerting: Basic threshold-based alerting for key metrics.
• Cross-Platform: Designed to run on Linux and Windows.

Prerequisites

• Python 3.10+
• Git (for cloning the repository)
• GPU Drivers: Ensure the appropriate drivers for your GPU(s) are installed.
  • NVIDIA: NVIDIA drivers, which include nvidia-smi.
  • AMD (Linux): AMDGPU drivers and the rocm-smi tool.
  • Google TPUs (Linux): TPU drivers (gasket, apex modules).
  • Windows: Standard WDDM drivers for all vendors.

Installation

First, clone the repository to your local machine:

git clone https://github.com/RohitRathore1/gpu-hardware-profiling-and-monitoring.git
cd gpu-hardware-profiling-and-monitoring

Next, install the required Python packages using pip. It is highly recommended to do this within a virtual environment.

# Create and activate a virtual environment (optional but recommended)
python3 -m venv venv
source venv/bin/activate  # On Windows, use `venv\Scripts\activate`

# Install dependencies
pip install -r requirements.txt

Usage

The tool is operated via the run.py script, which provides two main commands: profile and monitor.

To Run a System Profile:

This command performs a one-time scan of the system and prints a detailed JSON report of the hardware and software configuration.

python3 run.py profile

This is useful for quickly inventorying a machine's capabilities.

To Launch the Monitoring Dashboard:

This command starts a continuous monitoring service and launches a web server to display the data.

python3 run.py monitor

The dashboard will be available at http://127.0.0.1:8000.