Offline Transcription GPU Requirements (and the "no kernel image is available" Error)
Captyne's Offline / Local transcription runs Whisper Large v3.1 on your own graphics card using NVIDIA CUDA. On Windows this needs a reasonably modern NVIDIA GPU.
Supported GPUs (Windows)
You need an NVIDIA GPU with CUDA compute capability 7.5 or higher. That covers:
- GeForce RTX 20, 30, 40 and 50 series (RTX 2060 and up)
- GeForce GTX 16 series (GTX 1650, 1660 and their Super/Ti versions)
- Matching professional cards: Quadro RTX, RTX A-series and RTX Ada
You'll also need an up-to-date NVIDIA driver. Get the latest Game Ready or Studio driver from NVIDIA's website or the NVIDIA app.
Not supported
- GeForce GTX 10 series (GTX 1050, 1060, 1070, 1080 and 1080 Ti) and older cards
- AMD and Intel graphics on Windows
- Computers without a dedicated NVIDIA GPU
On these machines Captyne turns off the Offline option and tells you why. Older versions tried to run it anyway and failed with an error like:
CUDA error: no kernel image is available for execution on the device
That message means the GPU is too old for the CUDA version Captyne's offline engine is built with. Reinstalling won't fix it, and neither will the driver unless the driver itself is very old.
What to use instead
Your captions don't have to suffer. Switch the provider on the Setup tab to:
- Managed API. No setup, works on any PC, and uses your plan's monthly minutes.
- OpenAI, Groq or AssemblyAI with your own API key.
All of them give word-level timing, so the generated captions are the same. See Offline vs cloud transcription.
How to check your GPU
- Right-click the Start button → Device Manager → Display adapters and note the NVIDIA model.
- Or open Command Prompt and run
nvidia-smi. The GPU name is shown at the top.
Mac
On Macs, offline transcription is supported on Apple Silicon (M1 and newer). No separate GPU is needed.