Any applications with image processing are usually bottlenecked by the available RAM on a system. If you're running any ML applications, it's most efficient to run models on batches of images and batch size is determined by RAM.