It can fit larger models, but does a decent sdk exist for integration with current tools? How fast is it compared to an Nvidia card?