By consuming lots of RAM (more than will fit on a single Si die), it means effectively that the defender is able to leverage the economies of scale that go into optimizing the memory bus of his commodity server. This makes it so an attacker who can produce his own chips (or to a lesser extent, use FPGAs or GPUs) has much less of an advantage over the defender.
While CPU speeds, transistor densities, and cache sizes have gone through the roof, the 60-80 ns memory latency of off-chip DRAM has been nearly constant over the last few decades of computing.
While CPU speeds, transistor densities, and cache sizes have gone through the roof, the 60-80 ns memory latency of off-chip DRAM has been nearly constant over the last few decades of computing.
"Memory-hard" builds the work factor on that.