How much VRAM does Whisper need?
Whisper is far smaller than a chat model, so the weights dominate and the cache barely registers. The interesting number is how much faster than real time it runs.
The short answer
Computed from this tool’s default settings — whisper model and the rest as most people start. Change them below for your own case.
This Whisper size needs about 2.54 GB and fits RTX 4090 easily. Expect roughly 11× real time, so 60 minutes of audio in around 5.5 minutes.
Whisper VRAM Calculator
Memory and real-time factor for each Whisper size, on your card.
This Whisper size needs about 2.54 GB and fits RTX 4090 easily. Expect roughly 11× real time, so 60 minutes of audio in around 5.5 minutes.
| Model | Parameters | Needs | On your card |
|---|---|---|---|
| tiny | 39M | 1.08 GB | Fits |
| base | 74M | 1.15 GB | Fits |
| small | 244M | 1.49 GB | Fits |
| medium | 769M | 2.54 GB | Fits |
| large | 1,550M | 4.10 GB | Fits |
What each setting changes
Every input moves the result for a reason. This is what each one does and where to find the value for your own machine.
| Setting | Default | What it changes |
|---|---|---|
| Whisper model | medium — 769M | 5 options, from tiny — 39M to large — 1.55B. |
| Precision | float16 | 4 options, from float32 to 4-bit. |
| Your hardware | RTX 4090 · 24 GB | The machine the model runs on. Usable memory decides what fits and memory bandwidth decides how fast it runs, so this moves every figure below. |
| Audio to transcribe | 60 minutes | Your own figure in minutes, starting from 60. Change it to match what you actually run. |
Real answers across whisper model
The same calculation run at a range of settings, with everything else left at its default. These are computed by the tool itself, not written by hand.
| Whisper model | Fits | Weights | Your memory | Real-time factor |
|---|---|---|---|---|
| tiny — 39M | 1.08 GB | 0.08 GB | 22.1 GB | 215× |
| base — 74M | 1.15 GB | 0.15 GB | 22.1 GB | 114× |
| small — 244M | 1.49 GB | 0.49 GB | 22.1 GB | 34× |
| medium — 769M | 2.54 GB | 1.54 GB | 22.1 GB | 11× |
| large — 1.55B | 4.10 GB | 3.10 GB | 22.1 GB | 5× |
This Whisper size needs about 1.08 GB and fits RTX 4090 easily. Expect roughly 215× real time, so 60 minutes of audio in around 0.3 minutes.
How this is calculated
No lookup tables and no invented constants. Here is the arithmetic, so you can check it against your own numbers.
Whisper is an encoder-decoder model and much smaller than a chat LLM, so weights dominate and the cache is negligible. Memory is parameters × precision ÷ 8, plus roughly a gigabyte of runtime and audio buffers.
Throughput is quoted as a real-time factor — how many minutes of audio are processed per minute of wall clock. It is estimated from memory bandwidth in the same way as text generation, so treat it as an order of magnitude rather than a benchmark.
These are well-founded engineering estimates, not benchmark results. Your quantisation, runtime and context length all move the real number, and usable memory is an assumption rather than a specification. See the full methodology for every assumption behind these figures.
Who this is for
The three situations that bring people to this calculation.
Transcribing
Pick a model size your card handles.
Batch jobs
Estimate wall clock for hours of audio.
On a small card
Find the largest Whisper that fits.
How to use this calculator
Four steps, no account, nothing leaves your browser.
Set whisper model
Start at the top of the panel. Every figure recalculates as you change it — there is no submit button, because watching the number move is the point.
Adjust the rest to match your setup
3 further settings: precision, your hardware, audio to transcribe. Defaults are the common case, so change only what differs for you.
Read the headline, then the table
The large figure answers the question. The table underneath shows how the answer changes across nearby settings, which is usually where the decision actually gets made.
Check it against the method
The arithmetic is written out above. If a number looks wrong for your hardware, the assumptions are the first place to look — usable memory and quantisation are the two that vary most.
How much VRAM does Whisper need: common questions
The questions people ask about this, answered without hedging.
How much VRAM does Whisper need?
Large is about 1.55B parameters, so roughly 3 GB at float16 plus a gigabyte of overhead. Medium is around half that, and the smaller sizes run almost anywhere.
Which Whisper model should I use?
Large for accuracy and accented or noisy audio; small or medium is often enough for clean English speech and runs several times faster.
Can I run Whisper on CPU?
Yes — whisper.cpp handles it well, especially at the smaller sizes. Expect slower than real time on large, which is fine for batch transcription.
What is a real-time factor?
How many minutes of audio are processed per minute of wall clock. A factor of 10 means an hour of audio transcribes in six minutes.
Related local AI tools
All 50 run on the same arithmetic, so answers across them agree.
Will it fit
Context
Speed
Quantisation
Runtime & errors
Cost
Fine-tuning
Searches this page answers
Different ways of asking the same question, all resolved above.
Now find the models that fit
Sizing is only half the problem. Model Radar takes your hardware and shows which models actually run on it, ranked by what they are good at — the same arithmetic as this page, applied to every model worth running.