Whisper is OpenAI's open model for transcribing speech in 99 languages. The gpu.nz template runs a web interface and an API: upload a recording and get the text or SRT and VTT subtitles. Recognition runs on the GPU, so an hour of audio usually takes a few minutes.
In short
Whisper runs on cards with 8 GB or more. The most affordable suitable card right now is the RTX 3060 12 GB from 6.95 ₽ per hour, billed by the minute.
Updated 13 October 2026
The turbo and medium models fit into 8 GB of video memory. large-v3, the most accurate model, needs about 10 GB, so for it take a card with 12 GB or more, for example an RTX 3060 12 GB. For big queues of recordings a faster card finishes sooner, for example an RTX 4090 24 GB.
Prices and availability come from the live catalog and update automatically.
| Model | Video memory | When to pick it |
|---|---|---|
| tiny, base | about 1 GB | a quick check of the audio, a rough draft |
| small | about 2 GB | a fast draft on a weaker card |
| medium | about 5 GB | good quality on ordinary speech |
| turbo | about 6 GB | faster than large-v3 with a small loss in accuracy |
| large-v3 | about 10 GB | top accuracy: noisy recordings, rare languages, jargon |
The memory figures come from the Whisper model descriptions. For ordinary speech, turbo often gives a good balance of speed and accuracy; for important recordings take large-v3.
SRT is a common subtitle format: YouTube, video editors and most players read it. VTT is similar to SRT, but browsers and the video tag read it. A piece of an SRT file looks like this:
1
00:00:01,200 --> 00:00:04,800
Welcome to the lecture.
2
00:00:05,100 --> 00:00:09,300
Today we will look at how transformers work.If you need only the text without timestamps, copy it from the result. If you have many recordings, take the results through the API: send a file and save the response.
The API button on the server page opens the documentation at /docs, which lists every method and request field. Call transcription from your own script, and save the results to the server's disk or copy them over SSH. To connect, see the guide SSH access to a GPU server.