Comment by JSR_FDED

MacBooks with their unified memory behave like a slow GPU with enormous amount of video RAM. So you can run large smart models slowly.

Dedicated GPUs have less video RAM so can run smaller less smart models quickly.