Transcription resource estimator
Use this estimator to calculate the required CPU and GPU hours based on the total duration of audio to be processed.
Estimated resource usage
| Resource | Hours needed |
|---|---|
| CPU | – |
| GPU | – |
How are these values calculated?
- CPU hours = 32 × total audio hours
- GPU hours = 3.6% of total audio hours (for partial GPU usage - MIG), rounded up
Dictaphone users:
This estimate can also be used for Dictaphone. This is because the estimate is extra conservative. Dictaphone users are recommended to record audio with a machine type that does not use GPU resources.
How is the estimate conservative?
- There are enough resources suggested that you could transcribe your audio on either the CPU or GPU options.
- Both resources are also doubled. This is to allow you to attempt more than one transcriber model size if you don't get great results with the default settings.
Dictaphone users can launch the application with a small CPU machine type for recording audio, and later return to the recording data and transcribe this with larger CPU or GPU machine types.
Use this estimate as a guideline when requesting resources. If your workload is experimental or your audio material varies significantly, consider applying for slightly more resources than the estimate suggests.