GPU rendering helps for scaling images, rendering text and drawing other primitives efficiently. It frees up CPU time, and can redraw many times faster in scrolls and resizes. This is why most browsers use the GPU when most web content is 2D. GPU's don't really crunch huge numbers but instead utilise parallelisation.
The point of the browser-less claim is that it converts the md to html the same way most existing engines do, but then does not use a browser engine to render it, instead using something faster and more purpose-built.
As for the GPU-thing I'm not sure, but for example terminals are sometimes written to take advantage of GPU's (kitty for example: https://sw.kovidgoyal.net/kitty/#design-philosophy ) so why not this?