Generally CUDA is proprietary and only available for Nvidia hardware. However there's an open-source project called Zluda, which aims to improve CUDA compatibility with other gpu manufacturers: github.com/vosen/ZLUDA I'm not sure if they support arc cards or the cuda features pytorch uses yet, but it is still actively developed and maintained. If not, the CPU fallback should still give you managable speeds if you stay with the lower models.
So inspiring, mind blown once again!
Will it work with ARC Cards?
Generally CUDA is proprietary and only available for Nvidia hardware.
However there's an open-source project called Zluda, which aims to improve CUDA compatibility with other gpu manufacturers:
github.com/vosen/ZLUDA
I'm not sure if they support arc cards or the cuda features pytorch uses yet, but it is still actively developed and maintained.
If not, the CPU fallback should still give you managable speeds if you stay with the lower models.