What is GPU acceleration?
Using a computer's graphics card (GPU) to do heavy math, like running a speech model, much faster than the main processor alone.
GPU acceleration means handing heavy work to the graphics card. A GPU has thousands of small cores built for doing lots of math at once. That suits AI models, which are mostly big piles of multiplication.
Why it matters for transcription
A speech model turns audio into text by running millions of calculations per second of sound. On a GPU, that can be many times faster than on the main processor (the CPU). Faster means a live transcript that keeps up, and room for bigger, more accurate models.
What to look for
- An NVIDIA card is the most widely supported for AI work.
- Video memory (VRAM) decides how big a model fits.
- No GPU? Smaller, compressed models still run well on a modern CPU.
An example
Take a one-hour recording. On the main processor alone, a larger, more accurate speech model can take a while to work through it. On an NVIDIA graphics card, the same model often finishes many times faster, and can keep up live while people talk. The actual speed depends on the model, the card and the rest of the computer.
Common questions
- Do I need a gaming PC? No. A GPU helps, but transcription works without one.
- What about built-in graphics or AMD cards? Support varies by app and model. Many AI tools are built first for NVIDIA cards, so those are the safe bet.
- What about โAI PCsโ with an NPU? An NPU is a small chip for light AI tasks that saves battery. Many speech tools donโt use it yet, so check before you count on it.
- Will it slow my computer down? Transcription uses the GPU in bursts. You may hear the fans, but other work carries on.
Related: speech-to-text model and local transcription.
In Fetch
Fetch checks your computer during setup. With an NVIDIA card it uses it for faster, more accurate transcripts. Without one, it picks a model sized for your processor and memory. See System requirements and Choose a speech model.
Questions people ask
Do I need a graphics card to use Fetch?
No. Without one, your processor handles live notes. A graphics card makes transcripts faster and lets Fetch use bigger, more accurate models.
How much video memory do I need for transcription?
It depends on the model. Many speech models run comfortably with 4 to 8 GB of video memory, and the largest want more. With less, a compressed model or the main processor still works.
Still scratching your head?
Send us a note and a real person will get back to you.