> ## Documentation Index
> Fetch the complete documentation index at: https://docs.vowen.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Performance

> Fixing CPU, memory, and speed issues.

## Expected Resource Usage

| Scenario               | RAM Usage    | CPU             |
| ---------------------- | ------------ | --------------- |
| Idle (no model loaded) | \~200-300 MB | Near 0%         |
| Small model loaded     | \~400-600 MB | Near 0% (idle)  |
| Large v3 model loaded  | \~3-4 GB     | Near 0% (idle)  |
| During transcription   | +50-200 MB   | 30-100% (brief) |

<Note>RAM usage depends on which model is loaded. Models are loaded on first transcription and stay in memory for fast subsequent use.</Note>

## Transcription Is Slow

### On Windows (No GPU)

Local models can be slow on Windows without GPU acceleration. Options:

1. **Use a cloud model** — Groq offers fast, free transcription
2. **Enable GPU acceleration** — Settings > Models > Download GPU module (NVIDIA only)
3. **Use a smaller model** — Tiny or Base instead of Medium/Large
4. **Enable Resource Efficient Mode** — Settings > General (at bottom)

### On macOS

Local models are generally fast on Apple Silicon. If you're experiencing slowness:

1. **First transcription is always slower** — the model needs to load into memory
2. **Large v3** is naturally slower — try Large v3 Turbo for better speed
3. **Parakeet** is very fast on macOS — try switching to it
4. Check Activity Monitor for other apps consuming CPU

### With Cloud Models

If cloud transcription is slow:

* Check your internet connection
* The provider may be experiencing load — try a different one
* Gemini Flash Lite has been reported as occasionally slower during peak times

## High Memory Usage

### Why Does Vowen Use So Much RAM?

The transcription model stays loaded in memory between uses for faster response times. This is by design.

### Reducing Memory Usage

1. **Switch to a smaller model** — Base.en uses \~200 MB vs Large v3 at \~3 GB
2. **Enable Resource Efficient Mode** — loads the model only when needed (slower but less RAM)
3. **Use cloud models** — no local model needs to stay in memory

## MacBook Running Warm

Normal during active transcription — the model is using CPU/GPU intensively. This stops immediately when transcription is complete. Between recordings, Vowen should be idle.

If your MacBook runs warm even when not transcribing:

* Check Activity Monitor for Vowen's CPU usage
* Restart Vowen to reset any stuck processes
* Report the issue on Discord with your model configuration

## Battery Impact

Vowen has minimal battery impact when idle. During transcription, it briefly uses significant CPU, but recordings typically last only seconds.

**Tips for laptop users:**

* Use cloud models to avoid CPU-intensive local processing
* Enable Resource Efficient Mode to unload models when not in use
* Smaller models use less power per transcription