Performance
GPU Accelerated Voice Dictation
Leverage your GPU for faster speech recognition. Vocalinux supports AMD, Intel, and NVIDIA GPUs via Vulkan - with automatic fallback to CPU when needed.
Quick Check: Do You Have GPU Support?
Run this command to check if your system supports Vulkan GPU acceleration:
vulkaninfo --summary | grep deviceNameIf you see your GPU name, you're ready for accelerated dictation! On multi-GPU machines Vocalinux prefers a discrete Vulkan device when one is present; pick a specific device under Advanced settings if you need to override auto-select.
Supported GPU Backends
Vulkan
AMD, Intel, NVIDIA
Cross-platform GPU API supported by all major GPU vendors. The default and most compatible backend for whisper.cpp.
- Works with AMD, Intel, and NVIDIA GPUs
- No vendor-specific drivers required
- Excellent Linux support
- Best option for AMD and Intel GPUs
CUDA
NVIDIA only
NVIDIA's proprietary GPU compute platform. High performance but requires NVIDIA GPU and CUDA toolkit.
- Excellent performance on NVIDIA hardware
- Mature ecosystem and tooling
- Works with PyTorch-based Whisper engine
CPU
All systems
Pure CPU processing. Slower but works on any system. Automatic fallback when GPU is unavailable.
- Works on any hardware
- No GPU requirements
- Predictable performance
- Lowest power consumption option
Model Performance Comparison
Approximate transcription times for 5 seconds of audio. GPU times assume a mid-range Vulkan-compatible GPU.
| Model | Size | CPU Time | GPU Time | Accuracy | Best For |
|---|---|---|---|---|---|
| tiny | ~74MB | ~0.5s | ~0.2s | Good for real-time | Real-time dictation, older hardware |
| base | ~141MB | ~0.8s | ~0.3s | Better accuracy | Balanced speed/accuracy |
| small | ~465MB | ~1.5s | ~0.5s | High accuracy | Most users, good balance |
| medium | ~1.5GB | ~4s | ~1.2s | Very high accuracy | Accuracy-focused use, with GPU |
| large | ~3.0GB | ~8s | ~2.5s | Best accuracy | Maximum accuracy, requires GPU |
Hardware Requirements
Minimum (CPU-only)
- RAM
- 4GB RAM
- Storage
- 200MB
- GPU
- None required
Functional but slower processing
Recommended (with GPU)
- RAM
- 8GB RAM
- Storage
- 500MB
- GPU
- Any Vulkan-compatible GPU
Real-time dictation with small/base models
Optimal
- RAM
- 16GB+ RAM
- Storage
- 2GB+
- GPU
- NVIDIA RTX or AMD RX series
Real-time with any model including large
Intel GPU Note
Some Intel integrated GPUs don't fully support the compute features needed for whisper.cpp GPU acceleration. Vocalinux v0.7.0+ automatically detects incompatible Intel GPUs and falls back to CPU processing for reliability.
Get Started with GPU Acceleration
The installer automatically detects your GPU and configures the best backend. Just run and go.