Skip to content

GPU Acceleration

VoicePad uses your NVIDIA GPU to run Whisper transcription significantly faster than CPU. On an RTX 3050 (4 GB VRAM), the turbo model transcribes a 60-second recording in under 3 seconds.

NVIDIA GPUs only

GPU acceleration has been tested exclusively with NVIDIA GPUs. AMD and Apple Silicon GPUs are not supported.

What You Need

  • An NVIDIA GPU (GeForce, Quadro, or Tesla)
  • An up-to-date NVIDIA driver (version 520 or newer)

CUDA runtime libraries are bundled with VoicePad automatically. You do not need to install the CUDA toolkit, cuDNN, or any NVIDIA developer tools.

GPU VRAM Recommended Model
RTX 3050, RTX 4060 4 GB turbo
RTX 3060, RTX 4070 6-8 GB turbo or large-v3
RTX 3080, RTX 4090 10+ GB large-v3

Verifying GPU Is Active

When VoicePad starts, the header bar shows the active device:

model: turbo  device: cuda

If it shows device: cpu, VoicePad did not detect a usable NVIDIA GPU and fell back to CPU automatically.

Performance

Measured on RTX 3050 (4 GB VRAM), turbo model:

Audio Duration Transcription Time
10 seconds ~0.5 s
60 seconds ~2-3 s
5 minutes ~10-15 s

GPU acceleration provides roughly 4-10x speedup over CPU.

Troubleshooting

VoicePad shows device: cpu but I have an NVIDIA GPU

  1. Check your NVIDIA driver version by running nvidia-smi in a terminal
  2. Make sure the driver is version 520 or newer
  3. Reinstall VoicePad: uv tool install voicepad --reinstall

Out of memory error

Switch to a smaller model from the Settings tab. Try small or distil-small.en if your GPU has limited VRAM.