By GraphicCardPrices. Requirements checked against the official Whisper documentation on .
This guide is for podcasters, journalists, students and developers who want to turn recordings into text on their own computer, so interviews and meetings never get uploaded to a transcription service.
Whisper is OpenAI's open speech recognition model. You give it an audio file and it writes a transcript, and the multilingual models can also translate speech into English. It comes in six sizes, from tiny (39 million parameters) to large (1,550 million), plus turbo, an optimized version of large-v3.
License note: the README says Whisper's code and model weights are released under the MIT License, which permits commercial use.
What GPU do you need?
The official README says: "Below are the names of the available models and their approximate memory requirements and inference speed relative to the large model." (source)
| Tier | VRAM | What it runs | Example cards |
|---|---|---|---|
| Minimum | ~2 GB | The small model (244 million parameters) at about 2 GB. The tiny and base models need about 1 GB. source | RTX 3050, RTX 4060, RTX 3060 (12GB version) |
| Comfortable | ~6 GB | The turbo model, the README default, at about 6 GB. The medium model needs about 5 GB. source | RTX 4060, RTX 5060, RTX 3060 (12GB version) |
| Best | ~10 GB | The large model at about 10 GB. For translating into English, the README recommends medium or large rather than turbo. source | RTX 3060 (12GB version), RTX 4070 SUPER, RTX 5070 |
Lowest recorded prices by tier
The cheapest New and Used listings we have recorded for each tier, among NVIDIA GeForce RTX cards with at least that much VRAM. Used includes renewed and refurbished cards. Each is a recorded Amazon US price, not a live checkout price; the model page shows every listing and its condition.
| Tier | Lowest New | Lowest Used |
|---|---|---|
| Minimum (2GB+) | $249.99 RTX 3050 6GB Recorded 2026-10-27 22:20:12 UTC | $199.97 RTX 2060 6GB Recorded 2026-09-29 16:25:24 UTC |
| Comfortable (6GB+) | $249.99 RTX 3050 6GB Recorded 2026-10-27 22:20:12 UTC | $199.97 RTX 2060 6GB Recorded 2026-09-29 16:25:24 UTC |
| Best (10GB+) | $459.99 RTX 3060 12GB Recorded 2026-10-27 18:04:27 UTC | $409.99 RTX 3060 12GB Recorded 2026-10-27 12:12:27 UTC |
How to get started
- Install Whisper with
pip install -U openai-whisper. It also needs the command-line tool ffmpeg, which the README shows how to install with apt, pacman, Homebrew, Chocolatey or Scoop. - Transcribe a file from the terminal with
whisper audio.mp3 --model turbo. Swapturboforsmallormediumto match your card. - For English only, the README says the
.enmodels such assmall.entend to do better, especially at the tiny and base sizes, with a smaller difference at small and medium. - From Python,
whisper.load_model("turbo")loads the model and itstranscribemethod returns the text.
Full instructions are in the official Whisper repository.
Things to check before you buy
Translation needs a different model
The README says turbo is not trained for translation. If you ask turbo to translate it returns the original language, so use the medium or large model when you need English output from other languages.
NVIDIA, AMD and Intel
The README table gives memory figures without naming a GPU brand. Whisper is built on PyTorch, so check PyTorch's documentation for AMD or Intel support. The price table above covers NVIDIA GeForce RTX cards.
Speed
The README lists relative speed against the large model, measured on an A100, with turbo at about 8x and tiny at about 10x. It warns that real-world speed varies a lot with language, speaking speed and hardware. We have not benchmarked Whisper on consumer cards.
Which size should you choose?
Bigger models need more memory and run slower. Pick the largest one your card holds comfortably, and try small or turbo on a short clip first. The README gives no accuracy figures in the table, so judge quality on your own audio.
Whisper questions
How much VRAM does Whisper need?
The official README lists about 1 GB for tiny and base, about 2 GB for small, about 5 GB for medium, about 6 GB for turbo and about 10 GB for large. These are approximate figures.
Which Whisper model fits an 8GB graphics card?
By the README figures, turbo at about 6 GB and medium at about 5 GB both fit in 8GB. The large model at about 10 GB does not.
Is Whisper free for commercial use?
The README says the code and model weights are released under the MIT License, which allows commercial use. Make sure you have permission to transcribe the recordings you process.
Compare cards
Estimate memory for language models with the LLM VRAM estimator, see typical prices per memory size in GPU prices by VRAM, or compare pairs: RTX 3060 vs RTX 4060, RTX 5060 Ti vs RTX 4060 Ti. More local AI GPU guides.
Sources
- Whisper README (model sizes, required VRAM, install, license), checked
- Whisper license (MIT), checked
We have not benchmarked these tools. Requirements change between releases, so check the repository before you buy.