# Run Whisper large-v3

Turns speech into accurate text, in about a hundred languages. Ready in about 90 seconds on an RTX 4090, from $0.45/hr. No install, no code.

## What it does

Whisper listens to audio or video and writes down what was said, with timestamps.

It handles accents, background noise, crosstalk, and phone-quality recordings better than anything else available, and it works in roughly a hundred languages without being told which one it is hearing.

It will also translate as it goes, so a Spanish interview can come back as English text in one pass.

## Your setup

We run Whisper large-v3 on an RTX 4090 with 24GB of memory, at $0.45 per hour, billed by the minute. It takes about 2 minutes per hour of audio.

Whisper large-v3 only needs about 10GB of memory, so a cheaper card can run it. The RTX 4090 wins on throughput: it processes many segments of audio in parallel, and the batch size you can push through it makes a transcription job roughly three times faster than on an RTX 3090 for twice the hourly price. Faster and cheaper overall.

## What it costs

40 hours of audio takes about 80 minutes of GPU time, which is about $0.60.

## Two ways to run it

- **Open the app.** Drag a file in and read the transcript as it appears, with timestamps you can click.
- **Run a batch job.** The right mode for an archive. Connect a drive, transcribe everything, collect text and subtitle files.

## Model facts

| Field | Value |
|---|---|
| Parameters | 1.55B |
| License | Apache 2.0 |
| Memory required | 10GB minimum |
| Base model | Trained from scratch by OpenAI |
| Publisher | OpenAI |
| Recommended hardware | RTX 4090, 24GB |
| Category | [Voice and audio](https://gpuvault.io/working-on/voice-and-audio/) |

## Questions people ask

### How much does it cost to transcribe 40 hours of audio?

About $0.60. Forty hours of audio takes roughly 80 minutes on an RTX 4090 at $0.45 per hour. Commercial transcription services charge between $0.25 and $1.50 per minute of audio, which would put the same job between $600 and $3,600.

### Will it know who is speaking?

Not on its own. Whisper writes down what was said but does not separate speakers. Pair it with pyannote speaker diarization, which labels each segment by speaker, and run both in the same job. The transcription preset in the Job Runner does this for you when you tick "identify speakers".

### Can I use the transcripts commercially?

Yes. Whisper is Apache 2.0 licensed with no restrictions on commercial use or on what you do with the output.

### Do I need to install anything to use Whisper large-v3?

No. We start a machine with Whisper large-v3 already loaded and hand you a link. Everything runs in your browser, there is nothing to download, and nothing is left on your computer afterwards. It works the same on a Mac, a Windows laptop, or a Chromebook. A workspace is usually ready in about 90 seconds.

### How much does it cost to run 40 hours of audio?

About $0.60. Whisper large-v3 takes about 2 minutes per hour of audio on the RTX 4090 we recommend, so 40 hours of audio is roughly 80 minutes of GPU time at $0.45 per hour. Billing is by the minute, and a batch job shuts the GPU off the moment the last item finishes. The calculator above works this out for your own numbers.

### Can I run Whisper large-v3 on a cheaper card?

Sometimes, and the calculator will not always make it look worth it. Whisper large-v3 only needs about 10GB of memory, so a cheaper card can run it. The RTX 4090 wins on throughput: it processes many segments of audio in parallel, and the batch size you can push through it makes a transcription job roughly three times faster than on an RTX 3090 for twice the hourly price. Faster and cheaper overall. If you want to try a different card anyway, the advanced catalog lets you pick one and shows the estimated time before you commit.

---

*Source: https://gpuvault.io/model/whisper-large-v3/ — GPUVault, GPU rental by the minute.*