> not that it matters since it seems the GPU where most of the AI workload is handled anyway

GPUs can also have AI-specific compute (e.g., Nvidia’s tensor cores.)