- Messages
- 9,199
Training is done on CUDA or RT cores when you have an NVIDIA GPU, and it is done on the integrated GPU on Apple Silicon. But Apple Silicon just does not compete to the raw power of a dedicated GPU; so I'm not surprised at all that it still takes people around 30minutes on a non-dedicated card.
If you want an idea of the raw processing difference, you could do worse than look and compare using the Blender benchmarks:
opendata.blender.org
It really isn't surprising that Apple are being trounced when it comes to training models.
If you want an idea of the raw processing difference, you could do worse than look and compare using the Blender benchmarks:

Blender - Open Data
Blender Open Data is a platform to collect, display and query the results of hardware and software performance tests - provided by the public.

It really isn't surprising that Apple are being trounced when it comes to training models.