It also allows one to tune things to make the ratio of GPU parallelism to on-GPU memory to be sub-optimal. Making pure hardware implementations less practical

It also allows one to tune things to make the ratio of GPU parallelism to on-GPU memory to be sub-optimal. Making pure hardware implementations less practical isn't the only use of memory-hard work functions.