This is not entirely true.  There are very significant hardware architecture differences between CPUs and GPUs.  CPUs typically only have a few cores, but you can do anything you want with them.  GPUs have many cores on them (thousands), but they are optimized for running large numbers of pipelined vectorized instructions in parallel.  GPUs also have much less cache per core, and memory bandwidth between the GPU and the rest of the system is also an important consideration if you want to keep all of those cores busy.  This is why developers that want to try to make their coin "ASIC-resistant" or "GPU-resistant" use algorithms that require a lot of memory.