I wonder how they compare the machine learning performance of the integrated GPU. Does anyone know if any of the ML libraries support the new architecture? AFAI

I wonder how they compare the machine learning performance of the integrated GPU. Does anyone know if any of the ML libraries support the new architecture? AFAIK most libraries don't support anything other than nVidia. Any tests on training models? Or is it mostly inference?