GPU-SMI (Windows)

A lightweight Windows GPU monitoring tool based on Windows Performance Counters and WMI. No vendor CLI tools required.

Features

• ✅ Display GPU engine utilization
• ✅ Display dedicated/shared/total GPU memory usage
• ✅ Linux SMI-like output style
• ✅ Support all Windows GPUs (AMD, NVIDIA, Intel, etc.)
• ✅ No external dependencies, uses only Python standard library
• ✅ Continuous monitoring mode
• ✅ Verbose mode with process information

Requirements

• Windows 10/11 or Windows Server 2016+
• Python 3.6+
• PowerShell (built-in on Windows)

Installation

No additional dependencies required. Simply download gpu-smi.py and run.

# Clone or download the project
git clone <repository-url>
cd gpu-smi

# Or download the single file directly
curl -O gpu-smi.py

Usage

Basic Usage

# Show all GPUs
python gpu-smi.py

# Show a specific GPU (index starts from 0)
python gpu-smi.py -i 0

# Refresh every 5 seconds
python gpu-smi.py -l 5

# Show detailed information (including process list)
python gpu-smi.py -v

# Combine options
python gpu-smi.py -i 0 -l 2 -v

Command Line Arguments

Argument	Description
-i, --gpu INDEX	Specify GPU index to display (default: all)
-l, --loop SECONDS	Set refresh interval in seconds
-v, --verbose	Show detailed information (including GPU process list)
--version	Display version information
-h, --help	Show help message

Output Example

======================================================================
GPU-SMI (Windows) - 2025-03-08 14:30:15
======================================================================

======================================================================
GPU 0: AMD Radeon RX 6800 XT
======================================================================

GPU Memory:
  Dedicated GPU Memory:
    Total:      16.00 GiB
    Used:       8.52 GiB (53.3%)
    Available:  7.48 GiB
    [####################--------------------] 53.3%
  Shared GPU Memory:
    Total:      16.00 GiB
    Used:       2.15 GiB (13.4%)
    Available:  13.85 GiB
  Total GPU Memory:
    Total:      32.00 GiB
    Used:       10.67 GiB (33.3%)
    Available:  21.33 GiB
    [#############---------------------------] 33.3%

GPU Engine Utilization:
  3D        :  45.2%
  Copy      :   2.1%
  Compute   :  78.5%
  VideoEncode:  12.3%

Technical Implementation

This tool uses the following Windows native APIs to retrieve GPU information:

• WMI (Win32_VideoController): GPU device information and basic properties
• Performance Counters:
  • GPU Engine: Engine utilization
  • GPU Adapter Memory: Dedicated/shared memory usage
  • GPU Local/Non-Local Adapter Memory: Local/non-local memory usage

Memory Statistics

The tool provides three memory views:

1. Dedicated GPU Memory: GPU's dedicated VRAM
2. Shared GPU Memory: System memory shared with the GPU
3. Total GPU Memory: Total available GPU memory (dedicated + shared)

For integrated GPUs (UMA architecture), total memory may match the physical system memory.

Notes

1. Permissions: Administrator privileges may be required to access certain performance counters
2. WMI AdapterRAM Limitation: For VRAM over 4GB, WMI's AdapterRAM field may be inaccurate (uint32 limitation)
3. Integrated GPUs: For UMA devices, memory statistics may include shared system memory
4. Performance Overhead: The tool uses PowerShell queries; refresh interval should not be less than 1 second

Known Limitations

• Process-level GPU usage monitoring is a simplified implementation with limited accuracy
• Some older Windows versions or GPU drivers may not support all performance counters
• Temperature and power consumption not yet implemented (requires vendor-specific APIs)

Comparison with Vendor Tools

Feature	gpu-smi.py (this tool)	AMD SMI	NVIDIA SMI
Cross-vendor Support	✅ All GPUs	❌ AMD only	❌ NVIDIA only
External Dependencies	❌ None	✅ Required	✅ Required
Temperature Monitoring	❌	✅	✅
Power Monitoring	❌	✅	✅
Process-level Precision	Simplified	✅	✅
Windows Native	✅	❌	✅

Troubleshooting

Issue: "No GPU devices found"

Solutions:

• Ensure the system has physical GPU devices
• Check if GPU drivers are correctly installed
• Try running with administrator privileges

Issue: Memory shows 0 or inaccurate values

Solutions:

• Some counters require the GPU to be in use to display values
• Check if Windows Performance Counter service is running
• Review the notes in the output for data source information

Issue: PowerShell execution errors

Solutions:

• Ensure PowerShell is available (powershell -Command "Get-Date")
• Check PowerShell execution policy settings

Contributing

Issues and Pull Requests are welcome!

License

MIT License

Acknowledgments

Inspired by Linux's nvidia-smi and AMD's rocm-smi tools.