"I want to use Whisper, but I don't have a GPU" is a common concern. The short answer: Whisper works without a GPU. However, processing speed will be significantly slower.
This article covers how to run Whisper on CPU and tips for making it practical.
GPU vs CPU: Speed Differences
Why GPUs Are Faster
Deep learning models like Whisper perform massive parallel matrix operations. GPUs have thousands of cores for parallel processing, making them dramatically faster than CPUs with only dozens of cores.
Speed Benchmarks
Approximate processing times for a 10-minute audio file:
| Model | GPU (RTX 3060) | CPU (Core i7) |
|---|---|---|
| tiny | ~5 sec | ~30 sec |
| base | ~8 sec | ~1 min |
| small | ~20 sec | ~4 min |
| medium | ~50 sec | ~15 min |
| large-v3-turbo | ~20 sec | ~10 min |
| large-v3 | ~2 min | ~30+ min |
Actual speeds vary by CPU generation, core count, and available memory.
Running Whisper on CPU
Option 1: Official OpenAI Whisper
pip install openai-whisper
# Runs on CPU automatically when no GPU is detected
whisper audio.mp3 --model small --language en
Option 2: faster-whisper (Recommended)
faster-whisper is optimized with CTranslate2 and runs faster than official Whisper even on CPU:
pip install faster-whisper
from faster_whisper import WhisperModel
# Run on CPU with int8 quantization
model = WhisperModel("small", device="cpu", compute_type="int8")
segments, info = model.transcribe("audio.mp3", language="en")
Key: compute_type="int8" quantizes the model to 8-bit integers, improving CPU speed and reducing memory usage.
Option 3: Use a GUI App
WhisperApp automatically switches to CPU execution when no GPU is detected. Model selection and downloads are handled through the GUI — no command-line knowledge needed.
Tips for CPU Transcription
1. Use Smaller Models
On CPU, small (244M) or below is the most practical choice. Medium works but takes considerably longer.
2. Enable int8 Quantization
faster-whisper's compute_type="int8" dramatically speeds up CPU processing with virtually no accuracy loss.
3. Split Long Audio
Break lengthy audio into segments to reduce memory usage.
4. Run in Background
CPU transcription can make your PC sluggish, so run it in the background and wait for completion.
Consider Adding a GPU
If you transcribe frequently, investing in a GPU pays off quickly:
| GPU | Price Range | VRAM | Best Model |
|---|---|---|---|
| RTX 3050 | ~$150 | 4-8GB | small / medium |
| RTX 4060 | ~$300 | 8GB | large-v3-turbo |
| RTX 4070 | ~$550 | 12GB | large-v3 |
If you have an Intel GPU, WhisperApp's OpenVINO backend can accelerate processing. Vulkan-compatible GPUs also provide some speedup.
Conclusion
Whisper works without a GPU. For CPU environments, the small model with faster-whisper (int8 quantization) is the most practical combination.
However, if you transcribe regularly, adding a GPU dramatically improves workflow efficiency. WhisperApp auto-detects your CPU/GPU environment and runs with optimal settings, so you can start transcribing immediately regardless of your PC specs.
