GPU and CUDA

GPU & CUDA

TypeWhisper ships an optional CUDA backend that runs the bundled whisper.cpp transcription engine on a compatible NVIDIA GPU. This page covers what it requires and what it currently applies to.

Requirements

To use the CUDA backend you need:

• An NVIDIA GPU and driver.
• Only two system libraries on disk: libcudart.so.12 and libcublas.so.12 (from the CUDA 12 runtime).

You do not need a full CUDA toolkit install. The rest of the CUDA runtime the engine needs ships with the bundled whisper.cpp engine.

If those libraries are missing, the Linux UI disables the relevant control and explains what's needed — see Troubleshooting.

Choosing the backend

The Dictation section has an acceleration selector with three choices:

Option	Behavior
Auto	Use CUDA when available, otherwise CPU.
CPU	Force CPU.
NVIDIA CUDA	Force the CUDA backend (only selectable when a GPU and the CUDA 12 libraries are present).

Restart to switch

The process is pinned to the first backend it loads. Switching between CPU and CUDA therefore takes effect only after an app restart: when the active backend differs from your selection, the status text reads "Restart TypeWhisper to apply."

Fix CUDA path

When an NVIDIA GPU and CUDA 12 libraries exist on disk but aren't visible to the running process, TypeWhisper shows a Fix CUDA path button. Clicking it preloads libcudart.so.12 and libcublas.so.12 into the process so the engine can find them without LD_LIBRARY_PATH. Restart TypeWhisper afterward to load CUDA.

Scope

CUDA currently applies to the bundled whisper.cpp engine. Other bundled local engines stay on CPU.

Related pages

• Transcription engines — the engines available, including bundled whisper.cpp.
• Dictation — the acceleration selector and model settings.
• Requirements — full list of Linux requirements and optional helpers.

---

Changelog

Date	Change
2026-06-17	Initial version.
2026-06-17	Added acceleration selector, restart-to-switch behavior, Fix CUDA path button, and clarified only two libs are required.