Maybe it depends on how you quantify greatness. I think that if you're really good at writing a fast matrix kernel, that's lovely, but if you're really good at making a matrix kernel which is very useful to lots of people, or even just to your team or company, _that's_ great.