Docker GPU Dashboard Extension

Real-time NVIDIA GPU metrics directly in Docker Desktop. Monitor GPU utilization, memory usage, temperature, and power draw without leaving your IDE.

Features

• Real-time monitoring — 2-second polling for all NVIDIA GPUs on the host
• Per-GPU metrics — utilization, memory, temperature, power, fan speed
• Historical charts — 2-minute rolling graphs for all metrics
• Color-coded alerts — green/yellow/red thresholds for quick visual scanning
• Mock mode — develop and demo without GPU hardware
• Zero configuration — detects GPUs automatically via NVML

Installation

docker extension install pmady/docker-gpu-dashboard-extension:latest

Or build from source:

make install

Requirements

• Docker Desktop 4.30+
• NVIDIA GPU with drivers installed (or use mock mode for development)
• NVIDIA Container Toolkit (for real GPU metrics)

Development

Quick start with mock data (no GPU needed)

# Install frontend dependencies
make ui-deps

# Run backend in mock mode
make backend-mock

# In another terminal, start frontend dev server
cd ui && npm run dev

# Enable hot reload in Docker Desktop
make dev

Build the extension

make build

Validate

make validate

Architecture

Docker Desktop ←→ Extension UI (React + Recharts)
                      ↕
                Extension Backend (Go)
                      ↕
                NVML Library → GPU 0, GPU 1, ...

• Frontend: React 18 + Recharts for live charts, Vite for bundling
• Backend: Go HTTP server on Unix socket, polls NVML every 2 seconds
• Communication: Docker Extension API (ddClient.extension.vm.service)

Metrics Collected

Metric	Source	Unit
GPU Utilization	nvml.DeviceGetUtilizationRates()	%
Memory Used/Total	nvml.DeviceGetMemoryInfo()	bytes
Memory Controller Utilization	nvml.DeviceGetUtilizationRates()	%
Temperature	nvml.DeviceGetTemperature()	Celsius
Power Draw	nvml.DeviceGetPowerUsage()	Watts
Power Limit	nvml.DeviceGetEnforcedPowerLimit()	Watts
Fan Speed	nvml.DeviceGetFanSpeed()	%
Driver Version	nvml.SystemGetDriverVersion()	string

Related Projects

• keda-gpu-scaler — KEDA external scaler for GPU autoscaling
• otel-gpu-receiver — OpenTelemetry collector receiver for GPU metrics

Contributing

Contributions welcome! Please open an issue or PR.

License

Apache 2.0

---

Built by Pavan Madduri — Senior Cloud Platform Engineer, CNCF Golden Kubestronaut