GPU Acceleration

Voicepad supports NVIDIA GPU-accelerated transcription for 4-5x faster processing compared to CPU-only mode. GPU acceleration is optional and works seamlessly with automatic CPU fallback.

Overview

GPU Mode: ~2-3 seconds per minute of audio (requires NVIDIA GPU + [gpu] extra)

CPU Mode: ~8-10 seconds per minute of audio (default, works everywhere)

Transcription automatically detects GPU availability and falls back to CPU if needed—no manual configuration required.

Key Features

• ✅ Zero global system changes — CUDA libraries isolated in your virtual environment
• ✅ Works with any NVIDIA GPU — GTX 10-series and newer supported
• ✅ Auto-fallback — Seamlessly uses CPU if GPU unavailable
• ✅ No CUDA Toolkit required — Everything provided by [gpu] extra
• ✅ Conflict-free — Doesn't interfere with system CUDA or other projects

Quick Links

• Quick Start — Install and verify GPU works in 5 minutes
• Troubleshooting — Fix common issues and check performance
• Advanced Setup — Hardware requirements, deep configuration, technical details

Installation

The simplest way to enable GPU acceleration:

pip install voicepad-core[gpu]

Then verify it works:

voicepad config system

Look for [OK] Compatible next to "faster-whisper GPU" in the output.

Next Steps

• Not sure if your GPU is supported? Start with Requirements
• Want to get GPU working? Go to Quick Start
• Having issues? Check Troubleshooting