Well, I am not an expert but something that can perform the same simple operations massivly parallel. Meaning, you want to avoid branching for example and there are lots of footguns, otherwise everything will be slow, or not working. And debugging is a pain.

I guess it would be more accurate to say that GPUs are an array of massively parallel SIMD processors, which are their own kind of parallelism. It's parallelism all the way down

Oh so why do you ask, if you know it (better)? ;)

What is optimised for parallelism, is not optimised for single thread performance, which is what one expects out of a CPU ..